Demonstration of the topic coherence pipeline in Gensim

Introduction

We will be using the u_mass and c_v coherence for two different LDA models: a "good" and a "bad" LDA model. The good LDA model will be trained over 50 iterations and the bad one for 1 iteration. Hence in theory, the good LDA model will be able come up with better or more human-understandable topics. Therefore the coherence measure output for the good LDA model should be more (better) than that for the bad LDA model. This is because, simply, the good LDA model usually comes up with better topics that are more human interpretable.

In [2]:
import numpy as np
import logging
try:
    import pyLDAvis.gensim
except ImportError:
    ValueError("SKIP: please install pyLDAvis")
    
import json
import warnings
warnings.filterwarnings('ignore')  # To ignore all warnings that arise here to enhance clarity

from gensim.models.coherencemodel import CoherenceModel
from gensim.models.ldamodel import LdaModel
from gensim.models.hdpmodel import HdpModel
from gensim.models.wrappers import LdaVowpalWabbit, LdaMallet
from gensim.corpora.dictionary import Dictionary
from numpy import array

Set up logging

In [2]:
logger = logging.getLogger()
logger.setLevel(logging.DEBUG)
logging.debug("test")

Set up corpus

As stated in table 2 from this paper, this corpus essentially has two classes of documents. First five are about human-computer interaction and the other four are about graphs. We will be setting up two LDA models. One with 50 iterations of training and the other with just 1. Hence the one with 50 iterations ("better" model) should be able to capture this underlying pattern of the corpus better than the "bad" LDA model. Therefore, in theory, our topic coherence for the good LDA model should be greater than the one for the bad LDA model.

In [3]:
texts = [['human', 'interface', 'computer'],
         ['survey', 'user', 'computer', 'system', 'response', 'time'],
         ['eps', 'user', 'interface', 'system'],
         ['system', 'human', 'system', 'eps'],
         ['user', 'response', 'time'],
         ['trees'],
         ['graph', 'trees'],
         ['graph', 'minors', 'trees'],
         ['graph', 'minors', 'survey']]
In [4]:
dictionary = Dictionary(texts)
corpus = [dictionary.doc2bow(text) for text in texts]

Set up two topic models

We'll be setting up two different LDA Topic models. A good one and bad one. To build a "good" topic model, we'll simply train it using more iterations than the bad one. Therefore the u_mass coherence should in theory be better for the good model than the bad one since it would be producing more "human-interpretable" topics.

In [5]:
goodLdaModel = LdaModel(corpus=corpus, id2word=dictionary, iterations=50, num_topics=2)
badLdaModel = LdaModel(corpus=corpus, id2word=dictionary, iterations=1, num_topics=2)

Using U_Mass Coherence

In [14]:
goodcm = CoherenceModel(model=goodLdaModel, corpus=corpus, dictionary=dictionary, coherence='u_mass')
In [15]:
badcm = CoherenceModel(model=badLdaModel, corpus=corpus, dictionary=dictionary, coherence='u_mass')

View the pipeline parameters for one coherence model

Following are the pipeline parameters for u_mass coherence. By pipeline parameters, we mean the functions being used to calculate segmentation, probability estimation, confirmation measure and aggregation as shown in figure 1 in this paper.

In [16]:
print goodcm

CoherenceModel(segmentation=<function s_one_pre at 0x7f663ae82f50>, probability estimation=<function p_boolean_document at 0x7f663ae8a2a8>, confirmation measure=<function log_conditional_probability at 0x7f663ae8a668>, aggregation=<function arithmetic_mean at 0x7f663ae8aa28>)

Interpreting the topics

As we will see below using LDA visualization, the better model comes up with two topics composed of the following words:

  1. goodLdaModel:
    • Topic 1: More weightage assigned to words such as "system", "user", "eps", "interface" etc which captures the first set of documents.
    • Topic 2: More weightage assigned to words such as "graph", "trees", "survey" which captures the topic in the second set of documents.
  2. badLdaModel:
    • Topic 1: More weightage assigned to words such as "system", "user", "trees", "graph" which doesn't make the topic clear enough.
    • Topic 2: More weightage assigned to words such as "system", "trees", "graph", "user" which is similar to the first topic. Hence both topics are not human-interpretable.

Therefore, the topic coherence for the goodLdaModel should be greater for this than the badLdaModel since the topics it comes up with are more human-interpretable. We will see this using u_mass and c_v topic coherence measures.

Visualize topic models

In [17]:
pyLDAvis.enable_notebook()
In [18]:
pyLDAvis.gensim.prepare(goodLdaModel, corpus, dictionary)
Out[18]:
In [19]:
pyLDAvis.gensim.prepare(badLdaModel, corpus, dictionary)
Out[19]:
In [20]:
print goodcm.get_coherence()

-14.0842451581
In [21]:
print badcm.get_coherence()

-14.4434307511

Using C_V coherence

In [25]:
goodcm = CoherenceModel(model=goodLdaModel, texts=texts, dictionary=dictionary, coherence='c_v')
In [26]:
badcm = CoherenceModel(model=badLdaModel, texts=texts, dictionary=dictionary, coherence='c_v')

Pipeline parameters for C_V coherence

In [27]:
print goodcm

CoherenceModel(segmentation=<function s_one_set at 0x7f663ae8a050>, probability estimation=<function p_boolean_sliding_window at 0x7f663ae8a320>, confirmation measure=<function cosine_similarity at 0x7f663ae8a938>, aggregation=<function arithmetic_mean at 0x7f663ae8aa28>)
In [28]:
print goodcm.get_coherence()

0.552164532134
In [29]:
print badcm.get_coherence()

0.5269189184

Support for wrappers

This API supports gensim's ldavowpalwabbit and ldamallet wrappers as input parameter to model.

In [5]:
model1 = LdaVowpalWabbit('/home/devashish/vw-8', corpus=corpus, num_topics=2, id2word=dictionary, passes=50)
model2 = LdaVowpalWabbit('/home/devashish/vw-8', corpus=corpus, num_topics=2, id2word=dictionary, passes=1)
In [7]:
cm1 = CoherenceModel(model=model1, corpus=corpus, coherence='u_mass')
cm2 = CoherenceModel(model=model2, corpus=corpus, coherence='u_mass')
In [8]:
print cm1.get_coherence()
print cm2.get_coherence()

-14.075813889
-15.1740896045
In [20]:
model1 = LdaMallet('/home/devashish/mallet-2.0.8RC3/bin/mallet',corpus=corpus , num_topics=2, id2word=dictionary, iterations=50)
model2 = LdaMallet('/home/devashish/mallet-2.0.8RC3/bin/mallet',corpus=corpus , num_topics=2, id2word=dictionary, iterations=1)
In [21]:
cm1 = CoherenceModel(model=model1, texts=texts, coherence='c_v')
cm2 = CoherenceModel(model=model2, texts=texts, coherence='c_v')
In [22]:
print cm1.get_coherence()
print cm2.get_coherence()

0.581114877802
0.549865328265

Support for other topic models

The gensim topics coherence pipeline can be used with other topics models too. Only the tokenized topics should be made available for the pipeline. Eg. with the gensim HDP model

In [5]:
hm = HdpModel(corpus=corpus, id2word=dictionary)
In [8]:
# To get the topic words from the model
topics = []
for topic_id, topic in hm.show_topics(num_topics=10, formatted=False):
    topic = [word for word, _ in topic]
    topics.append(topic)
In [9]:
topics[:2]
Out[9]:
[[u'minors',
  u'system',
  u'graph',
  u'human',
  u'interface',
  u'eps',
  u'trees',
  u'computer',
  u'user',
  u'response',
  u'survey',
  u'time'],
 [u'minors',
  u'trees',
  u'time',
  u'interface',
  u'user',
  u'survey',
  u'system',
  u'response',
  u'human',
  u'computer',
  u'graph',
  u'eps']]
In [10]:
# Initialize CoherenceModel using `topics` parameter
cm = CoherenceModel(topics=topics, corpus=corpus, dictionary=dictionary, coherence='u_mass')
In [11]:
cm.get_coherence()
Out[11]:
-14.640667699204982

Conclusion

Hence as we can see, the u_mass and c_v coherence for the good LDA model is much more (better) than that for the bad LDA model. This is because, simply, the good LDA model usually comes up with better topics that are more human interpretable. The badLdaModel however fails to decipher between these two topics and comes up with topics which are not clear to a human. The u_mass and c_v topic coherences capture this wonderfully by giving the interpretability of these topics a number as we can see above. Hence this coherence measure can be used to compare difference topic models based on their human-interpretability.