Skip to content

KeyNMF

KeyNMF is a topic model that relies on contextually sensitive embeddings for keyword retrieval and term importance estimation, while taking inspiration from classical matrix-decomposition approaches for extracting topics.

Schematic overview of KeyNMF

Here's an example of how you can fit and interpret a KeyNMF model in the easiest way.

from turftopic import KeyNMF

model = KeyNMF(10, encoder="paraphrase-MiniLM-L3-v2")
model.fit(corpus)

model.print_topics()

Which Embedding model should I use

How does KeyNMF work?

Keyword Extraction

KeyNMF discovers topics based on the importances of keywords for a given document. This is done by embedding words in a document, and then extracting the cosine similarities of documents to words using a transformer-model. Only the top_n keywords with positive similarity are kept.

Click to see formula
    1. Let \(x_d\) be the document's embedding produced with the encoder model.
      1. Let \(v_w\) be the word's embedding produced with the encoder model.
      2. Calculate cosine similarity between word and document

      For each word \(w\) in the document \(d\):

      \[ \text{sim}(d, w) = \frac{x_d \cdot v_w}{||x_d|| \cdot ||v_w||} \]

    For each document \(d\):

    1. Let \(K_d\) be the set of \(N\) keywords with the highest cosine similarity to document \(d\).
    \[ K_d = \text{argmax}_{K^*} \sum_{w \in K^*}\text{sim}(d,w)\text{, where } |K_d| = N\text{, and } \\ w \in d \]

  • Arrange positive keyword similarities into a keyword matrix \(M\) where the rows represent documents, and columns represent unique keywords.

    \[ M_{dw} = \begin{cases} \text{sim}(d,w), & \text{if } w \in K_d \text{ and } \text{sim}(d,w) > 0 \\ 0, & \text{otherwise}. \end{cases} \]

You can do this step manually if you want to precompute the keyword matrix. Keywords are represented as dictionaries mapping words to keyword importances.

model.extract_keywords(["Cars are perhaps the most important invention of the last couple of centuries. They have revolutionized transportation in many ways."])

[{'transportation': 0.44713873,
  'invention': 0.560524,
  'cars': 0.5046208,
  'revolutionized': 0.3339205,
  'important': 0.21803442}]

A precomputed Keyword matrix can also be used to fit a model:

keyword_matrix = model.extract_keywords(corpus)
model.fit(None, keywords=keyword_matrix)

Topic Discovery

Topics in this matrix are then discovered using Non-negative Matrix Factorization. Essentially the model tries to discover underlying dimensions/factors along which most of the variance in term importance can be explained.

Click to see formula

  • Decompose \(M\) with non-negative matrix factorization: \(M \approx WH\), where \(W\) is the document-topic matrix, and \(H\) is the topic-term matrix. Non-negative Matrix Factorization is done with the coordinate-descent algorithm, minimizing square loss:

    \[ L(W,H) = ||M - WH||^2 \]

You can fit KeyNMF on the raw corpus, with precomputed embeddings or with precomputed keywords.

model.fit(corpus)

from sentence_transformers import SentenceTransformer

trf = SentenceTransformer("all-MiniLM-L6-v2")
embeddings = trf.encode(corpus)

model = KeyNMF(10, encoder=trf)
model.fit(corpus, embeddings=embeddings)

keyword_matrix = model.extract_keywords(corpus)
model.fit(None, keywords=keyword_matrix)

Seeded Topic Modeling

When investigating a set of documents, you might already have an idea about what aspects you would like to explore. In KeyNMF, you can describe this aspect, from which you want to investigate your corpus, using a free-text seed-phrase, which will then be used to only extract topics, which are relevant to your research question.

How is this done?

KeyNMF encodes the seed phrase into a seed-embedding. Word importance scores in a document get weighted by their similarity to the seed-embedding.

  • Embed seed-phrase into a seed-embedding: \(s\)
  • When extracting keywords from a document:
    1. Let \(x_d\) be the document's embedding produced with the encoder model.
    2. Let the document's relevance be \(r_d = \text{sim}(d,w)\)
    3. For each word \(w\):
      1. Let the word's importance in the keyword matrix be: \(\text{sim}(d, w) \cdot r_d\) if \(r_d > 0\), otherwise \(0\)
from turftopic import KeyNMF

model = KeyNMF(5, seed_phrase="<your seed phrase>")
model.fit(corpus)

model.print_topics()
Topic ID Highest Ranking
0 morality, moral, immoral, morals, objective, morally, animals, society, species, behavior
1 armenian, armenians, genocide, armenia, turkish, turks, soviet, massacre, azerbaijan, kurdish
2 murder, punishment, death, innocent, penalty, kill, crime, moral, criminals, executed
3 gun, guns, firearms, crime, handgun, firearm, weapons, handguns, law, criminals
4 jews, israeli, israel, god, jewish, christians, sin, christian, palestinians, christianity
Topic ID Highest Ranking
0 atheist, atheists, religion, religious, theists, beliefs, christianity, christian, religions, agnostic
1 bible, christians, christian, christianity, church, scripture, religion, jesus, faith, biblical
2 god, existence, exist, exists, universe, creation, argument, creator, believe, life
3 believe, faith, belief, evidence, blindly, believing, gods, believed, beliefs, convince
4 atheism, atheists, agnosticism, belief, arguments, believe, existence, alt, believing, argument
Topic ID Highest Ranking
0 windows, dos, os, microsoft, ms, apps, pc, nt, file, shareware
1 ram, motherboard, card, monitor, memory, cpu, vga, mhz, bios, intel
2 unix, os, linux, intel, systems, programming, applications, compiler, software, platform
3 disk, scsi, disks, drive, floppy, drives, dos, controller, cd, boot
4 software, mac, hardware, ibm, graphics, apple, computer, pc, modem, program

Dynamic Topic Modeling

KeyNMF is also capable of modeling topics over time. This happens by fitting a KeyNMF model first on the entire corpus, then fitting individual topic-term matrices using coordinate descent based on the document-topic and document-term matrices in the given time slices.

Click to see formula
  1. Compute keyword matrix \(M\) for the whole corpus.
  2. Decompose \(M\) with non-negative matrix factorization: \(M \approx WH\).
    1. Let \(W_t\) be the document-topic proportions for documents in time slice \(t\), and \(M_t\) be the keyword matrix for words in time slice \(t\).
    2. Obtain the topic-term matrix for the time slice, by minimizing square loss using coordinate descent and fixing \(W_t\):

    For each time slice \(t\):

    \[ H_t = \text{argmin}_{H^{*}} ||M_t - W_t H^{*}||^2 \]

Here's an example of using KeyNMF in a dynamic modeling setting:

from datetime import datetime

from turftopic import KeyNMF

corpus: list[str] = []
timestamps: list[datetime] = []

model = KeyNMF(5, top_n=5, random_state=42)
document_topic_matrix = model.fit_transform_dynamic(
    corpus, timestamps=timestamps, bins=10
)

You can use the print_topics_over_time() method for producing a table of the topics over the generated time slices.

This example uses CNN news data.

model.print_topics_over_time()
| Time Slice | 0_olympics_tokyo_athletes_beijing | 1_covid_vaccine_pandemic_coronavirus | 2_olympic_athletes_ioc_athlete | 3_djokovic_novak_tennis_federer | 4_ronaldo_cristiano_messi_manchester | | - | - | - | - | - | - | | 2012 12 06 - 2013 11 10 | genocide, yugoslavia, karadzic, facts, cnn | cnn, russia, chechnya, prince, merkel | france, cnn, francois, hollande, bike | tennis, tournament, wimbledon, grass, courts | beckham, soccer, retired, david, learn | | 2013 11 10 - 2014 10 14 | keith, stones, richards, musician, author | georgia, russia, conflict, 2008, cnn | civil, rights, hear, why, should | cnn, kidneys, traffickers, organ, nepal | ronaldo, cristiano, goalscorer, soccer, player | | | | ... | | | | | 2020 05 07 - 2021 04 10 | olympics, beijing, xinjiang, ioc, boycott | covid, vaccine, coronavirus, pandemic, vaccination | olympic, japan, medalist, canceled, tokyo | djokovic, novak, tennis, federer, masterclass | ronaldo, cristiano, messi, juventus, barcelona | | 2021 04 10 - 2022 03 16 | olympics, tokyo, athletes, beijing, medal | covid, pandemic, vaccine, vaccinated, coronavirus | olympic, athletes, ioc, medal, athlete | djokovic, novak, tennis, wimbledon, federer | ronaldo, cristiano, messi, manchester, scored |

You can also display the topics over time on an interactive HTML figure. The most important words for topics get revealed by hovering over them.

You will need to install Plotly for this to work.

pip install plotly

model.plot_topics_over_time()
Topics over time in a Dynamic KeyNMF model.

Hierarchical Topic Modeling

When you suspect that subtopics might be present in the topics you find with the model, KeyNMF can be used to discover topics further down the hierarchy.

This is done by utilising a special case of weighted NMF, where documents are weighted by how high they score on the parent topic.

Click to see formula
  1. Decompose keyword matrix \(M \approx WH\)
  2. To find subtopics in topic \(j\), define document weights \(w\) as the \(j\)th column of \(W\).
  3. Estimate subcomponents with wNMF \(M \approx \mathring{W} \mathring{H}\) with document weight \(w\)
    1. Initialise \(\mathring{H}\) and \(\mathring{W}\) randomly.
    2. Perform multiplicative updates until convergence.
      \(\mathring{W}^T = \mathring{W}^T \odot \frac{\mathring{H} \cdot (M^T \odot w)}{\mathring{H} \cdot \mathring{H}^T \cdot (\mathring{W}^T \odot w)}\)
      \(\mathring{H}^T = \mathring{H}^T \odot \frac{ (M^T \odot w)\cdot \mathring{W}}{\mathring{H}^T \cdot (\mathring{W}^T \odot w) \cdot \mathring{W}}\)
  4. To sufficiently differentiate the subcomponents from each other a pseudo-c-tf-idf weighting scheme is applied to \(\mathring{H}\):
    1. \(\mathring{H} = \mathring{H}_{ij} \odot ln(1 + \frac{A}{1+\sum_k \mathring{H}_{kj}})\), where \(A\) is the average of all elements in \(\mathring{H}\)

To create a hierarchical model, you can use the hierarchy property of the model.

# This divides each of the topics in the model to 3 subtopics.
model.hierarchy.divide_children(n_subtopics=3)
print(model.hierarchy)
Root
├── 0: windows, dos, os, disk, card, drivers, file, pc, files, microsoft
│ ├── 0.0: dos, file, disk, files, program, windows, disks, shareware, norton, memory
│ ├── 0.1: os, unix, windows, microsoft, apps, nt, ibm, ms, os2, platform
│ └── 0.2: card, drivers, monitor, driver, vga, ram, motherboard, cards, graphics, ati
└── 1: atheism, atheist, atheists, religion, christians, religious, belief, christian, god, beliefs
. ├── 1.0: atheism, alt, newsgroup, reading, faq, islam, questions, read, newsgroups, readers
. ├── 1.1: atheists, atheist, belief, theists, beliefs, religious, religion, agnostic, gods, religions
. └── 1.2: morality, bible, christian, christians, moral, christianity, biblical, immoral, god, religion

For a detailed tutorial on hierarchical modeling click here.

Cross-lingual KeyNMF

KeyNMF, by default, does not come with cross-lingual capabilities, since only words that appear in a document can be assigned to it as keywords. We, however provide a term-matching scheme that allows you to match words across languages based on their cosine similarity in a multilingual embedding model.

This is done by:

  1. Computing a similarity matrix over terms.
  2. Checking, which terms have similarity over a given threshold (0.9 is the default)
  3. Building a graph from these connections, and finding graph components.
  4. Adding up term importances for terms that appear in the same component for all documents.
from datasets import load_dataset
from sklearn.feature_extraction.text import CountVectorizer

from turftopic import KeyNMF

# Loading a parallel corpus
ds = load_dataset(
    "aiana94/polynews-parallel", "deu_Latn-eng_Latn", split="train"
)
# Subsampling
ds = ds.train_test_split(test_size=1000)["test"]
corpus = ds["src"] + ds["tgt"]

model = KeyNMF(
    10,
    cross_lingual=True,
    encoder="paraphrase-multilingual-MiniLM-L12-v2",
    vectorizer=CountVectorizer()
)
model.fit(corpus)
model.print_topics()
Topic ID Highest Ranking
...
15 africa-afrikanisch-african, media-medien-medienwirksam, schwarzwald-negroe-schwarzer, apartheid, difficulties-complicated-problems, kontinents-continent-kontinent, äthiopien-ethiopia, investitionen-investiert-investierenden, satire-satirical, hundred-100-1001
16 lawmaker-judges-gesetzliche, schutz-sicherheit-geschützt, an-success-eintreten, australian-australien-australischen, appeal-appealing-appeals, lawyer-lawyers-attorney, regeln-rule-rules, öffentlichkeit-öffentliche-publicly, terrorism-terroristischer-terrorismus, convicted
17 israels-israel-israeli, palästinensischen-palestinians-palestine, gay-lgbtq-gays, david, blockaden-blockades-blockade, stars-star-stelle, aviv, bombardieren-bombenexplosion-bombing, militärischer-army-military, kampfflugzeuge-warplanes
18 russischer-russlands-russischen, facebookbeitrag-facebook-facebooks, soziale-gesellschaftliche-sozialbauten, internetnutzer-internet, activism-aktivisten-activists, webseiten-web-site, isis, netzwerken-networks-netzwerk, vkontakte, media-medien-medienwirksam
19 bundesstaates-regierenden-regiert, chinesischen-chinesische-chinesisch, präsidentschaft-presidential-president, regions-region-regionen, demokratien-democratic-democracy, kapitalismus-capitalist-capitalism, staatsbürgerin-citizens-bürger, jemen-jemenitische-yemen, angolanischen-angola, media-medien-medienwirksam

Online Topic Modeling

KeyNMF can also be fitted in an online manner. This is done by fitting NMF with batches of data instead of the whole dataset at once.

Use Cases:

  1. You can use online fitting when you have very large corpora at hand, and it would be impractical to fit a model on it at once.
  2. You have new data flowing in constantly, and need a model that can morph the topics based on the incoming data. You can also do this in a dynamic fashion.
  3. You need to finetune an already fitted topic model to novel data.

Batch Fitting

We will use the batching function from the itertools recipes to produce batches.

In newer versions of Python (>=3.12) you can just from itertools import batched

def batched(iterable, n: int):
    "Batch data into lists of length n. The last batch may be shorter."
    if n < 1:
        raise ValueError("n must be at least one")
    it = iter(iterable)
    while batch := tuple(itertools.islice(it, n)):
        yield batch

You can fit a KeyNMF model to a very large corpus in batches like so:

from turftopic import KeyNMF

model = KeyNMF(10, top_n=5)

corpus = ["some string", "etc", ...]
for batch in batched(corpus, 200):
    batch = list(batch)
    model.partial_fit(batch)

Precomputing the Keyword Matrix

If you desire the best results, it might make sense for you to go over the corpus in multiple epochs:

for epoch in range(5):
    for batch in batched(corpus, 200):
        model.partial_fit(batch)

This is mildly inefficient, however, as the texts need to be encoded on every epoch, and keywords need to be extracted. In such scenarios you might want to precompute and maybe even save the extracted keywords to disk using the extract_keywords() method.

Keywords are represented as dictionaries mapping words to keyword importances.

model.extract_keywords(["Cars are perhaps the most important invention of the last couple of centuries. They have revolutionized transportation in many ways."])

[{'transportation': 0.44713873,
  'invention': 0.560524,
  'cars': 0.5046208,
  'revolutionized': 0.3339205,
  'important': 0.21803442}]

You can extract keywords in batches and save them to disk to a file format of your choice. In this example I will use NDJSON because of its simplicity.

import json
from pathlib import Path
from typing import Iterable

# Here we are saving keywords to a JSONL/NDJSON file
with Path("keywords.jsonl").open("w") as keyword_file:
    # Doing this in batches is much more efficient than individual texts because
    # of the encoding.
    for batch in batched(corpus, 200):
        batch_keywords = model.extract_keywords(batch)
        # We serialize each
        for keywords in batch_keywords:
            keyword_file.write(json.dumps(keywords) + "\n")

def stream_keywords() -> Iterable[dict[str, float]]:
    """This function streams keywords from the file."""
    with Path("keywords.jsonl").open() as keyword_file:
        for line in keyword_file:
            yield json.loads(line.strip())

for epoch in range(5):
    keyword_stream = stream_keywords()
    for keyword_batch in batched(keyword_stream, 200):
        model.partial_fit(keywords=keyword_batch)

Dynamic Online Topic Modeling

KeyNMF can be online fitted in a dynamic manner as well. This is useful when you have large corpora of text over time, or when you want to fit the model on future information flowing in and want to analyze the topics' changes over time.

When using dynamic online topic modeling you have to predefine the time bins that you will use, as the model can't infer these from the data.

from datetime import datetime

# We will bin by years in a period of 2020-2030
bins = [datetime(year=y, month=1, day=1) for y in range(2020, 2030 + 2, 1)]

You can then online fit a dynamic topic model with partial_fit_dynamic().

model = KeyNMF(5, top_n=10)

corpus: list[str] = [...]
timestamps: list[datetime] = [...]

for batch in batched(zip(corpus, timestamps)):
    text_batch, ts_batch = zip(*batch)
    model.partial_fit_dynamic(text_batch, timestamps=ts_batch, bins=bins)

Asymmetric and Instruction-tuned Embedding Models

Some embedding models can be used together with prompting, or encode queries and passages differently. This is important for KeyNMF, as it is explicitly based on keyword retrieval, and its performance can be substantially enhanced by using asymmetric or prompted embeddings. Microsoft's E5 models are, for instance, all prompted by default, and it would be detrimental to performance not to do so yourself.

In these cases, you're better off NOT passing a string to Turftopic models, but explicitly loading the model using sentence-transformers.

Here's an example of using instruct models for keyword retrieval with KeyNMF. In this case, documents will serve as the queries and words as the passages:

from turftopic import KeyNMF
from sentence_transformers import SentenceTransformer

encoder = SentenceTransformer(
    "intfloat/multilingual-e5-large-instruct",
    prompts={
        "query": "Instruct: Retrieve relevant keywords from the given document. Query: "
        "passage": "Passage: "
    },
    # Make sure to set default prompt to query!
    default_prompt_name="query",
)
model = KeyNMF(10, encoder=encoder)

And a regular, asymmetric example:

encoder = SentenceTransformer(
    "intfloat/e5-large-v2",
    prompts={
        "query": "query: "
        "passage": "passage: "
    },
    # Make sure to set default prompt to query!
    default_prompt_name="query",
)
model = KeyNMF(10, encoder=encoder)

Setting the default prompt to query is especially important, when you are precomputing embeddings, as query should always be your default prompt to embed documents with.

API Reference

turftopic.models.keynmf.KeyNMF

Bases: ContextualModel, DynamicTopicModel, MultimodalModel

Extracts keywords from documents based on semantic similarity of term encodings to document encodings. Topics are then extracted with non-negative matrix factorization from keywords' proximity to documents.

from turftopic import KeyNMF

corpus: list[str] = ["some text", "more text", ...]

model = KeyNMF(10, top_n=10).fit(corpus)
model.print_topics()

Parameters:

Name Type Description Default
n_components int

Number of topics.

 required
encoder Union[Encoder, str, MultimodalEncoder]

Model to encode documents/terms, all-MiniLM-L6-v2 is the default.

 'sentence-transformers/all-MiniLM-L6-v2'
vectorizer Optional[CountVectorizer]

Vectorizer used for term extraction. Can be used to prune or filter the vocabulary.

 None
top_n int

Number of keywords to extract for each document.

 25
random_state Optional[int]

Random state to use so that results are exactly reproducible.

 None
metric Literal['cosine', 'dot']

Similarity metric to use for keyword extraction.

 'cosine'
seed_phrase Optional[str]

Describes an aspect of the corpus that the model should explore. It can be a free-text query, such as "Christian Denominations: Protestantism and Catholicism"

 None
cross_lingual bool

Indicates whether KeyNMF should match terms across languages. This is useful when you have a corpus containing multiple languages.

 False
term_match_threshold float

Cosine similarity threshold for matching terms across languages.

 0.9
Source code in turftopic/models/keynmf.py 
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
class KeyNMF(ContextualModel, DynamicTopicModel, MultimodalModel):
    """Extracts keywords from documents based on semantic similarity of
    term encodings to document encodings.
    Topics are then extracted with non-negative matrix factorization from
    keywords' proximity to documents.

    ```python
    from turftopic import KeyNMF

    corpus: list[str] = ["some text", "more text", ...]

    model = KeyNMF(10, top_n=10).fit(corpus)
    model.print_topics()
    ```

    Parameters
    ----------
    n_components: int
        Number of topics.
    encoder: str or SentenceTransformer
        Model to encode documents/terms, all-MiniLM-L6-v2 is the default.
    vectorizer: CountVectorizer, default None
        Vectorizer used for term extraction.
        Can be used to prune or filter the vocabulary.
    top_n: int, default 25
        Number of keywords to extract for each document.
    random_state: int, default None
        Random state to use so that results are exactly reproducible.
    metric: "cosine" or "dot", default "cosine"
        Similarity metric to use for keyword extraction.
    seed_phrase: str, default None
        Describes an aspect of the corpus that the model should explore.
        It can be a free-text query, such as
        "Christian Denominations: Protestantism and Catholicism"
    cross_lingual: bool, default False
        Indicates whether KeyNMF should match terms across languages.
        This is useful when you have a corpus containing multiple languages.
    term_match_threshold: float, default 0.9
        Cosine similarity threshold for matching terms across languages.
    """

    def __init__(
        self,
        n_components: int,
        encoder: Union[
            Encoder, str, MultimodalEncoder
        ] = "sentence-transformers/all-MiniLM-L6-v2",
        vectorizer: Optional[CountVectorizer] = None,
        top_n: int = 25,
        random_state: Optional[int] = None,
        metric: Literal["cosine", "dot"] = "cosine",
        seed_phrase: Optional[str] = None,
        cross_lingual: bool = False,
        term_match_threshold: float = 0.9,
    ):
        self.random_state = random_state
        self.n_components = n_components
        self.top_n = top_n
        self.metric = metric
        self.encoder = encoder
        self._has_custom_vectorizer = vectorizer is not None
        if isinstance(encoder, str):
            if (
                encoder == "sentence-transformers/all-MiniLM-L6-v2"
            ) and cross_lingual:
                warnings.warn(
                    "all-MiniLM is incompatible with cross-lingual transfer, using paraphrase-multilingual-MiniLM-L12-v2."
                )
                encoder = "paraphrase-multilingual-MiniLM-L12-v2"
            self.encoder_ = SentenceTransformer(encoder)
        else:
            self.encoder_ = encoder
        self.validate_encoder()
        if vectorizer is None:
            self.vectorizer = default_vectorizer()
        else:
            self.vectorizer = vectorizer
        self.model = KeywordNMF(
            n_components=n_components, seed=random_state, top_n=self.top_n
        )
        self.extractor = SBertKeywordExtractor(
            top_n=self.top_n,
            vectorizer=self.vectorizer,
            encoder=self.encoder_,
            metric=self.metric,
        )
        self.seed_phrase = seed_phrase
        self.seed_embedding = None
        if self.seed_phrase is not None:
            self.seed_embedding = self.encoder_.encode([self.seed_phrase])[0]
        self.cross_lingual = cross_lingual
        self.term_match_threshold = term_match_threshold

    def extract_keywords(
        self,
        batch_or_document: Union[str, list[str]],
        embeddings: Optional[np.ndarray] = None,
        fitting: bool = True,
    ) -> list[dict[str, float]]:
        """Extracts keywords from a document or a batch of documents.

        Parameters
        ----------
        batch_or_document: str | list[str]
            A single document or a batch of documents.
        embeddings: ndarray, optional
            Precomputed document embeddings.
        """
        if isinstance(batch_or_document, str):
            batch_or_document = [batch_or_document]
        keywords = self.extractor.batch_extract_keywords(
            batch_or_document,
            embeddings=embeddings,
            seed_embedding=self.seed_embedding,
            fitting=fitting,
        )
        if self.cross_lingual:
            keywords = self.extractor.match_terms(
                keywords, threshold=self.term_match_threshold
            )
        return keywords

    def vectorize(
        self,
        raw_documents=None,
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
    ) -> spr.csr_array:
        """Creates document-term-matrix from documents."""
        if keywords is None:
            keywords = self.extract_keywords(
                raw_documents, embeddings=embeddings
            )
        return self.model.vectorize(keywords)

    def divide_topic(
        self,
        node: DivisibleTopicNode,
        n_subtopics: int,
    ) -> list[DivisibleTopicNode]:
        document_term_matrix = getattr(self, "document_term_matrix", None)
        if document_term_matrix is None:
            raise ValueError(
                "document_term_matrix is needed for computing hierarchies. Perhaps you fitted the model online?"
            )
        dtm = document_term_matrix
        subtopics = []
        weight = node.document_topic_vector
        subcomponents, sub_doc_topic = weighted_nmf(
            dtm, weight, n_subtopics, self.random_state, max_iter=200
        )
        subcomponents = subcomponents * np.log(
            1 + subcomponents.mean() / (subcomponents.sum(axis=0) + 1)
        )
        subcomponents = normalize(subcomponents, axis=1, norm="l2")
        for i, component, doc_topic_vector in zip(
            range(n_subtopics), subcomponents, sub_doc_topic.T
        ):
            sub = DivisibleTopicNode(
                self,
                path=(*node.path, i),
                word_importance=component,
                document_topic_vector=doc_topic_vector,
                children=None,
            )
            subtopics.append(sub)
        return subtopics

    def fit_transform(
        self,
        raw_documents=None,
        y=None,
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
    ) -> np.ndarray:
        """Fits topic model and returns topic importances for documents.

        Parameters
        ----------
        raw_documents: iterable of str, optional
            Documents to fit the model on.
        embeddings: ndarray of shape (n_documents, n_dimensions), optional
            Precomputed document encodings.
        keywords: list[dict[str, float]], optional
            Precomputed keyword dictionaries.

        Returns
        -------
        ndarray of shape (n_dimensions, n_topics)
            Document-topic matrix.
        """
        if self.seed_phrase is not None and keywords is not None:
            warnings.warn(
                "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
            )
        console = Console()
        with console.status("Running KeyNMF") as status:
            if keywords is None:
                status.update("Extracting keywords")
                keywords = self.extract_keywords(
                    raw_documents, embeddings=embeddings
                )
                console.log("Keyword extraction done.")
            status.update("Decomposing with NMF")
            try:
                doc_topic_matrix = self.model.transform(keywords)
            except (NotFittedError, AttributeError):
                doc_topic_matrix = self.model.fit_transform(keywords)
                self.components_ = self.model.components
            console.log("Model fitting done.")
        self.document_topic_matrix = doc_topic_matrix
        self.document_term_matrix = self.model.vectorize(keywords)
        self.hierarchy = DivisibleTopicNode.create_root(
            self, self.components_, self.document_topic_matrix
        )
        return doc_topic_matrix

    def fit_transform_multimodal(
        self,
        raw_documents: list[str],
        images: list[ImageRepr],
        y=None,
        embeddings: Optional[MultimodalEmbeddings] = None,
    ) -> np.ndarray:
        self.validate_embeddings(embeddings)
        console = Console()
        self.multimodal_embeddings = embeddings
        with console.status("Fitting model") as status:
            if self.multimodal_embeddings is None:
                status.update("Encoding documents")
                self.multimodal_embeddings = self.encode_multimodal(
                    raw_documents, images
                )
                console.log("Documents encoded.")
            status.update("Extracting keywords")
            document_keywords = self.extract_keywords(
                raw_documents,
                embeddings=self.multimodal_embeddings["document_embeddings"],
            )
            image_keywords = self.extract_keywords(
                raw_documents,
                embeddings=self.multimodal_embeddings["image_embeddings"],
            )
            console.log("Keyword extraction done.")
            status.update("Decomposing with NMF")
            try:
                doc_topic_matrix = self.model.transform(document_keywords)
            except (NotFittedError, AttributeError):
                doc_topic_matrix = self.model.fit_transform(document_keywords)
                self.components_ = self.model.components
            console.log("Model fitting done.")
            status.update("Transforming images")
            self.image_topic_matrix = self.model.transform(image_keywords)
            self.top_images: list[list[Image.Image]] = self.collect_top_images(
                images, self.image_topic_matrix
            )
            console.log("Images transformed")
        self.document_topic_matrix = doc_topic_matrix
        self.document_term_matrix = self.model.vectorize(document_keywords)
        self.hierarchy = DivisibleTopicNode.create_root(
            self, self.components_, self.document_topic_matrix
        )
        return doc_topic_matrix

    def fit(
        self,
        raw_documents=None,
        y=None,
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
    ) -> np.ndarray:
        """Fits topic model and returns topic importances for documents.

        Parameters
        ----------
        raw_documents: iterable of str, optional
            Documents to fit the model on.
        embeddings: ndarray of shape (n_documents, n_dimensions), optional
            Precomputed document encodings.
        keywords: list[dict[str, float]], optional
            Precomputed keyword dictionaries.
        """
        self.fit_transform(
            raw_documents, y, embeddings=embeddings, keywords=keywords
        )
        return self

    def get_vocab(self) -> np.ndarray:
        return np.array(self.model.index_to_key)

    def transform(
        self,
        raw_documents=None,
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
    ) -> np.ndarray:
        """Infers topic importances for new documents based on a fitted model.

        Parameters
        ----------
        raw_documents: iterable of str
            Documents to fit the model on.
        embeddings: ndarray of shape (n_documents, n_dimensions), optional
            Precomputed document encodings.
        keywords: list[dict[str, float]], optional
            Precomputed keyword dictionaries.

        Returns
        -------
        ndarray of shape (n_dimensions, n_topics)
            Document-topic matrix.
        """
        if self.seed_phrase is not None and keywords is not None:
            warnings.warn(
                "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
            )
        if keywords is None and raw_documents is None:
            raise ValueError(
                "You have to pass either keywords or raw_documents."
            )
        if keywords is None:
            keywords = self.extract_keywords(
                list(raw_documents),
                embeddings=embeddings,
                fitting=False,
            )
        return self.model.transform(keywords)

    def partial_fit(
        self,
        raw_documents: Optional[list[str]] = None,
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
    ):
        """Online fits KeyNMF on a batch of documents.

        Parameters
        ----------
        raw_documents: iterable of str
            Documents to fit the model on.
        embeddings: ndarray of shape (n_documents, n_dimensions), optional
            Precomputed document encodings.
        keywords: list[dict[str, float]], optional
            Precomputed keyword dictionaries.
        """
        if self.seed_phrase is not None and keywords is not None:
            warnings.warn(
                "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
            )
        if self.cross_lingual:
            raise ValueError(
                "Cross-lingual online topic modeling is yet to be implemented in KeyNMF."
            )
        if not self._has_custom_vectorizer:
            self.vectorizer = CountVectorizer(stop_words="english")
            self._has_custom_vectorizer = True
        min_df = self.vectorizer.min_df
        max_df = self.vectorizer.max_df
        if (min_df != 1) or (max_df != 1.0):
            warnings.warn(
                f"""When applying partial fitting, the vectorizer is fitted batch-wise in KeyNMF.
            You have a vectorizer with min_df={min_df}, and max_df={max_df}.
            If you continue with these settings, all tokens might get filtered out.
            We recommend setting min_df=1 and max_df=1.0 for online fitting.
            `model = KeyNMF(10, vectorizer=CountVectorizer(min_df=1, max_df=1.0)`
            """
            )
        if keywords is None and raw_documents is None:
            raise ValueError(
                "You have to pass either keywords or raw_documents."
            )
        if keywords is None:
            keywords = self.extract_keywords(
                raw_documents, embeddings=embeddings
            )
        self.model.partial_fit(keywords)
        self.components_ = self.model.components
        return self

    def prepare_topic_data(
        self,
        corpus: Optional[list[str]],
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
    ) -> TopicData:
        if keywords is None and corpus is None:
            raise ValueError(
                "You have to pass either keywords or raw_documents."
            )
        if self.seed_phrase is not None and keywords is not None:
            warnings.warn(
                "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
            )
        console = Console()
        with console.status("Running KeyNMF") as status:
            if embeddings is None:
                embeddings = self.encode_documents(corpus)
            if keywords is None:
                status.update("Extracting keywords")
                keywords = self.extract_keywords(corpus, embeddings=embeddings)
                console.log("Keyword extraction done.")
            if (corpus is not None) and (len(keywords) != len(corpus)):
                raise ValueError(
                    "length of keywords is not the same as length of the corpus"
                )
            status.update("Decomposing with NMF")
            try:
                doc_topic_matrix = self.model.transform(keywords)
            except (NotFittedError, AttributeError):
                doc_topic_matrix = self.model.fit_transform(keywords)
                self.components_ = self.model.components
            console.log("Model fitting done.")
            document_term_matrix = self.model.vectorize(keywords)
        self.document_topic_matrix = doc_topic_matrix
        self.document_term_matrix = document_term_matrix
        self.hierarchy = DivisibleTopicNode.create_root(
            self, self.components_, self.document_topic_matrix
        )
        res = TopicData(
            corpus=corpus,
            document_term_matrix=document_term_matrix,
            vocab=self.get_vocab(),
            document_topic_matrix=doc_topic_matrix,
            document_representation=embeddings,
            topic_term_matrix=self.components_,  # type: ignore
            transform=getattr(self, "transform", None),
            topic_names=self.topic_names,
            hierarchy=getattr(self, "hierarchy", None),
        )
        return res

    def prepare_dynamic_topic_data(
        self,
        corpus: Optional[list[str]],
        timestamps: list[datetime],
        embeddings: Optional[np.ndarray] = None,
        bins: Union[int, list[datetime]] = 10,
        keywords: Optional[list[dict[str, float]]] = None,
    ):
        if self.seed_phrase is not None and keywords is not None:
            warnings.warn(
                "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
            )
        if ((corpus is not None) and (keywords is not None)) and (
            len(keywords) != len(corpus)
        ):
            raise ValueError(
                "Length of keywords is not the same as length of the corpus"
            )
        if (corpus is None) and (keywords is None):
            raise TypeError(
                "You have to pass keywords or a corpus, but both are None."
            )
        if embeddings is None:
            embeddings = self.encode_documents(corpus)
        console = Console()
        with console.status("Running KeyNMF") as status:
            if embeddings is None:
                embeddings = self.encode_documents(corpus)
            if keywords is None:
                status.update("Extracting keywords")
                keywords = self.extract_keywords(corpus, embeddings=embeddings)
                console.log("Keyword extraction done.")
            if (corpus is not None) and (len(keywords) != len(corpus)):
                raise ValueError(
                    "length of keywords is not the same as length of the corpus"
                )
            status.update("Decomposing with NMF")
            try:
                doc_topic_matrix = self.model.transform(keywords)
            except (NotFittedError, AttributeError):
                doc_topic_matrix = self.fit_transform_dynamic(
                    corpus,
                    embeddings=embeddings,
                    keywords=keywords,
                    bins=bins,
                    timestamps=timestamps,
                )
            console.log("Model fitting done.")
            document_term_matrix = self.model.vectorize(keywords)
        try:
            classes = self.classes_
        except AttributeError:
            classes = list(range(self.components_.shape[0]))
        res = TopicData(
            corpus=corpus,
            document_term_matrix=document_term_matrix,
            vocab=self.get_vocab(),
            document_topic_matrix=doc_topic_matrix,
            document_representation=embeddings,
            topic_term_matrix=self.components_,  # type: ignore
            transform=getattr(self, "transform", None),
            topic_names=self.topic_names,
            classes=classes,
            temporal_components=self.temporal_components_,
            temporal_importance=self.temporal_importance_,
            time_bin_edges=self.time_bin_edges,
        )
        return res

    def fit_transform_dynamic(
        self,
        raw_documents=None,
        timestamps: Optional[list[datetime]] = None,
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
        bins: Union[int, list[datetime]] = 10,
    ) -> np.ndarray:
        if self.seed_phrase is not None and keywords is not None:
            warnings.warn(
                "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
            )
        if timestamps is None:
            raise TypeError(
                "You have to pass timestamps when fitting a dynamic model."
            )
        if keywords is None and raw_documents is None:
            raise ValueError(
                "You have to pass either keywords or raw_documents."
            )
        if keywords is None:
            keywords = self.extract_keywords(
                raw_documents, embeddings=embeddings
            )
        time_labels, self.time_bin_edges = self.bin_timestamps(
            timestamps, bins
        )
        doc_topic_matrix = self.model.fit_transform_dynamic(
            keywords, time_labels, self.time_bin_edges
        )
        self.temporal_importance_ = (
            self.model.temporal_importance_.T
            / self.model.temporal_importance_.sum(axis=1)
        ).T
        self.temporal_components_ = self.model.temporal_components
        self.components_ = self.model.components
        self.document_topic_matrix = doc_topic_matrix
        self.document_term_matrix = self.model.vectorize(keywords)
        self.hierarchy = DivisibleTopicNode.create_root(
            self, self.components_, self.document_topic_matrix
        )
        return doc_topic_matrix

    def partial_fit_dynamic(
        self,
        raw_documents=None,
        timestamps: Optional[list[datetime]] = None,
        embeddings: Optional[np.ndarray] = None,
        keywords: Optional[list[dict[str, float]]] = None,
        bins: Union[int, list[datetime]] = 10,
    ):
        """Online fits Dynamic KeyNMF on a batch of documents.

        Parameters
        ----------
        raw_documents: iterable of str
            Documents to fit the model on.
        embeddings: ndarray of shape (n_documents, n_dimensions), optional
            Precomputed document encodings.
        keywords: list[dict[str, float]], optional
            Precomputed keyword dictionaries.
        timestamps: list[datetime], optional
            List of timestamps for the batch.
        bins: list[datetime]
            Explicit time bin edges for the dynamic model.
        """
        if self.seed_phrase is not None and keywords is not None:
            warnings.warn(
                "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
            )
        if self.cross_lingual:
            raise ValueError(
                "Cross-lingual online topic modeling is yet to be implemented in KeyNMF."
            )
        if timestamps is None:
            raise TypeError(
                "You have to pass timestamps when fitting a dynamic model."
            )
        if keywords is None and raw_documents is None:
            raise ValueError(
                "You have to pass either keywords or raw_documents."
            )
        time_bin_edges = getattr(self, "time_bin_edges", None)
        if time_bin_edges is None:
            if isinstance(bins, int):
                raise TypeError(
                    "You have to pass explicit time bins (list of time bin edges) when partial "
                    "fitting KeyNMF, at least at the first call."
                )
            else:
                self.time_bin_edges = bins
        time_labels, self.time_bin_edges = self.bin_timestamps(
            timestamps, self.time_bin_edges
        )
        if keywords is None:
            keywords = self.extract_keywords(
                raw_documents, embeddings=embeddings
            )
        self.model.partial_fit_dynamic(
            keywords, time_labels, self.time_bin_edges
        )
        self.temporal_importance_ = (
            self.model.temporal_importance_.T
            / self.model.temporal_importance_.sum(axis=1)
        ).T
        self.temporal_components_ = self.model.temporal_components
        self.components_ = self.model.components
        return self

extract_keywords(batch_or_document, embeddings=None, fitting=True)

Extracts keywords from a document or a batch of documents.

Parameters:

Name Type Description Default
batch_or_document Union[str, list[str]]

A single document or a batch of documents.

 required
embeddings Optional[ndarray]

Precomputed document embeddings.

 None
Source code in turftopic/models/keynmf.py 
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
def extract_keywords(
    self,
    batch_or_document: Union[str, list[str]],
    embeddings: Optional[np.ndarray] = None,
    fitting: bool = True,
) -> list[dict[str, float]]:
    """Extracts keywords from a document or a batch of documents.

    Parameters
    ----------
    batch_or_document: str | list[str]
        A single document or a batch of documents.
    embeddings: ndarray, optional
        Precomputed document embeddings.
    """
    if isinstance(batch_or_document, str):
        batch_or_document = [batch_or_document]
    keywords = self.extractor.batch_extract_keywords(
        batch_or_document,
        embeddings=embeddings,
        seed_embedding=self.seed_embedding,
        fitting=fitting,
    )
    if self.cross_lingual:
        keywords = self.extractor.match_terms(
            keywords, threshold=self.term_match_threshold
        )
    return keywords

fit(raw_documents=None, y=None, embeddings=None, keywords=None)

Fits topic model and returns topic importances for documents.

Parameters:

Name Type Description Default
raw_documents

Documents to fit the model on.

 None
embeddings Optional[ndarray]

Precomputed document encodings.

 None
keywords Optional[list[dict[str, float]]]

Precomputed keyword dictionaries.

 None
Source code in turftopic/models/keynmf.py 
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
def fit(
    self,
    raw_documents=None,
    y=None,
    embeddings: Optional[np.ndarray] = None,
    keywords: Optional[list[dict[str, float]]] = None,
) -> np.ndarray:
    """Fits topic model and returns topic importances for documents.

    Parameters
    ----------
    raw_documents: iterable of str, optional
        Documents to fit the model on.
    embeddings: ndarray of shape (n_documents, n_dimensions), optional
        Precomputed document encodings.
    keywords: list[dict[str, float]], optional
        Precomputed keyword dictionaries.
    """
    self.fit_transform(
        raw_documents, y, embeddings=embeddings, keywords=keywords
    )
    return self

fit_transform(raw_documents=None, y=None, embeddings=None, keywords=None)

Fits topic model and returns topic importances for documents.

Parameters:

Name Type Description Default
raw_documents

Documents to fit the model on.

 None
embeddings Optional[ndarray]

Precomputed document encodings.

 None
keywords Optional[list[dict[str, float]]]

Precomputed keyword dictionaries.

 None

Returns:

Type Description
ndarray of shape (n_dimensions, n_topics)

Document-topic matrix.
Source code in turftopic/models/keynmf.py 
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
def fit_transform(
    self,
    raw_documents=None,
    y=None,
    embeddings: Optional[np.ndarray] = None,
    keywords: Optional[list[dict[str, float]]] = None,
) -> np.ndarray:
    """Fits topic model and returns topic importances for documents.

    Parameters
    ----------
    raw_documents: iterable of str, optional
        Documents to fit the model on.
    embeddings: ndarray of shape (n_documents, n_dimensions), optional
        Precomputed document encodings.
    keywords: list[dict[str, float]], optional
        Precomputed keyword dictionaries.

    Returns
    -------
    ndarray of shape (n_dimensions, n_topics)
        Document-topic matrix.
    """
    if self.seed_phrase is not None and keywords is not None:
        warnings.warn(
            "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
        )
    console = Console()
    with console.status("Running KeyNMF") as status:
        if keywords is None:
            status.update("Extracting keywords")
            keywords = self.extract_keywords(
                raw_documents, embeddings=embeddings
            )
            console.log("Keyword extraction done.")
        status.update("Decomposing with NMF")
        try:
            doc_topic_matrix = self.model.transform(keywords)
        except (NotFittedError, AttributeError):
            doc_topic_matrix = self.model.fit_transform(keywords)
            self.components_ = self.model.components
        console.log("Model fitting done.")
    self.document_topic_matrix = doc_topic_matrix
    self.document_term_matrix = self.model.vectorize(keywords)
    self.hierarchy = DivisibleTopicNode.create_root(
        self, self.components_, self.document_topic_matrix
    )
    return doc_topic_matrix

partial_fit(raw_documents=None, embeddings=None, keywords=None)

Online fits KeyNMF on a batch of documents.

Parameters:

Name Type Description Default
raw_documents Optional[list[str]]

Documents to fit the model on.

 None
embeddings Optional[ndarray]

Precomputed document encodings.

 None
keywords Optional[list[dict[str, float]]]

Precomputed keyword dictionaries.

 None
Source code in turftopic/models/keynmf.py 
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
def partial_fit(
    self,
    raw_documents: Optional[list[str]] = None,
    embeddings: Optional[np.ndarray] = None,
    keywords: Optional[list[dict[str, float]]] = None,
):
    """Online fits KeyNMF on a batch of documents.

    Parameters
    ----------
    raw_documents: iterable of str
        Documents to fit the model on.
    embeddings: ndarray of shape (n_documents, n_dimensions), optional
        Precomputed document encodings.
    keywords: list[dict[str, float]], optional
        Precomputed keyword dictionaries.
    """
    if self.seed_phrase is not None and keywords is not None:
        warnings.warn(
            "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
        )
    if self.cross_lingual:
        raise ValueError(
            "Cross-lingual online topic modeling is yet to be implemented in KeyNMF."
        )
    if not self._has_custom_vectorizer:
        self.vectorizer = CountVectorizer(stop_words="english")
        self._has_custom_vectorizer = True
    min_df = self.vectorizer.min_df
    max_df = self.vectorizer.max_df
    if (min_df != 1) or (max_df != 1.0):
        warnings.warn(
            f"""When applying partial fitting, the vectorizer is fitted batch-wise in KeyNMF.
        You have a vectorizer with min_df={min_df}, and max_df={max_df}.
        If you continue with these settings, all tokens might get filtered out.
        We recommend setting min_df=1 and max_df=1.0 for online fitting.
        `model = KeyNMF(10, vectorizer=CountVectorizer(min_df=1, max_df=1.0)`
        """
        )
    if keywords is None and raw_documents is None:
        raise ValueError(
            "You have to pass either keywords or raw_documents."
        )
    if keywords is None:
        keywords = self.extract_keywords(
            raw_documents, embeddings=embeddings
        )
    self.model.partial_fit(keywords)
    self.components_ = self.model.components
    return self

partial_fit_dynamic(raw_documents=None, timestamps=None, embeddings=None, keywords=None, bins=10)

Online fits Dynamic KeyNMF on a batch of documents.

Parameters:

Name Type Description Default
raw_documents

Documents to fit the model on.

 None
embeddings Optional[ndarray]

Precomputed document encodings.

 None
keywords Optional[list[dict[str, float]]]

Precomputed keyword dictionaries.

 None
timestamps Optional[list[datetime]]

List of timestamps for the batch.

 None
bins Union[int, list[datetime]]

Explicit time bin edges for the dynamic model.

 10
Source code in turftopic/models/keynmf.py 
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
def partial_fit_dynamic(
    self,
    raw_documents=None,
    timestamps: Optional[list[datetime]] = None,
    embeddings: Optional[np.ndarray] = None,
    keywords: Optional[list[dict[str, float]]] = None,
    bins: Union[int, list[datetime]] = 10,
):
    """Online fits Dynamic KeyNMF on a batch of documents.

    Parameters
    ----------
    raw_documents: iterable of str
        Documents to fit the model on.
    embeddings: ndarray of shape (n_documents, n_dimensions), optional
        Precomputed document encodings.
    keywords: list[dict[str, float]], optional
        Precomputed keyword dictionaries.
    timestamps: list[datetime], optional
        List of timestamps for the batch.
    bins: list[datetime]
        Explicit time bin edges for the dynamic model.
    """
    if self.seed_phrase is not None and keywords is not None:
        warnings.warn(
            "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
        )
    if self.cross_lingual:
        raise ValueError(
            "Cross-lingual online topic modeling is yet to be implemented in KeyNMF."
        )
    if timestamps is None:
        raise TypeError(
            "You have to pass timestamps when fitting a dynamic model."
        )
    if keywords is None and raw_documents is None:
        raise ValueError(
            "You have to pass either keywords or raw_documents."
        )
    time_bin_edges = getattr(self, "time_bin_edges", None)
    if time_bin_edges is None:
        if isinstance(bins, int):
            raise TypeError(
                "You have to pass explicit time bins (list of time bin edges) when partial "
                "fitting KeyNMF, at least at the first call."
            )
        else:
            self.time_bin_edges = bins
    time_labels, self.time_bin_edges = self.bin_timestamps(
        timestamps, self.time_bin_edges
    )
    if keywords is None:
        keywords = self.extract_keywords(
            raw_documents, embeddings=embeddings
        )
    self.model.partial_fit_dynamic(
        keywords, time_labels, self.time_bin_edges
    )
    self.temporal_importance_ = (
        self.model.temporal_importance_.T
        / self.model.temporal_importance_.sum(axis=1)
    ).T
    self.temporal_components_ = self.model.temporal_components
    self.components_ = self.model.components
    return self

transform(raw_documents=None, embeddings=None, keywords=None)

Infers topic importances for new documents based on a fitted model.

Parameters:

Name Type Description Default
raw_documents

Documents to fit the model on.

 None
embeddings Optional[ndarray]

Precomputed document encodings.

 None
keywords Optional[list[dict[str, float]]]

Precomputed keyword dictionaries.

 None

Returns:

Type Description
ndarray of shape (n_dimensions, n_topics)

Document-topic matrix.
Source code in turftopic/models/keynmf.py 
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
def transform(
    self,
    raw_documents=None,
    embeddings: Optional[np.ndarray] = None,
    keywords: Optional[list[dict[str, float]]] = None,
) -> np.ndarray:
    """Infers topic importances for new documents based on a fitted model.

    Parameters
    ----------
    raw_documents: iterable of str
        Documents to fit the model on.
    embeddings: ndarray of shape (n_documents, n_dimensions), optional
        Precomputed document encodings.
    keywords: list[dict[str, float]], optional
        Precomputed keyword dictionaries.

    Returns
    -------
    ndarray of shape (n_dimensions, n_topics)
        Document-topic matrix.
    """
    if self.seed_phrase is not None and keywords is not None:
        warnings.warn(
            "Seed phrase is specified, but keyword matrix is pre-computed. The seed phrase will be ignored. Note that this is not a problem if you already calculated the keyword matrix using the seed phrase."
        )
    if keywords is None and raw_documents is None:
        raise ValueError(
            "You have to pass either keywords or raw_documents."
        )
    if keywords is None:
        keywords = self.extract_keywords(
            list(raw_documents),
            embeddings=embeddings,
            fitting=False,
        )
    return self.model.transform(keywords)

vectorize(raw_documents=None, embeddings=None, keywords=None)

Creates document-term-matrix from documents.

Source code in turftopic/models/keynmf.py 
148
149
150
151
152
153
154
155
156
157
158
159
def vectorize(
    self,
    raw_documents=None,
    embeddings: Optional[np.ndarray] = None,
    keywords: Optional[list[dict[str, float]]] = None,
) -> spr.csr_array:
    """Creates document-term-matrix from documents."""
    if keywords is None:
        keywords = self.extract_keywords(
            raw_documents, embeddings=embeddings
        )
    return self.model.vectorize(keywords)