rJST {sentopics}

Create a Reversed Joint Sentiment/Topic model

Description

This function initialize a Reversed Joint Sentiment/Topic model.

Usage

rJST(x, ...)

## Default S3 method:
rJST(
  x,
  lexicon = NULL,
  K = 5,
  S = 3,
  alpha = 1,
  gamma = 5,
  beta = 0.01,
  alphaCycle = 0,
  gammaCycle = 0,
  ...
)

## S3 method for class 'LDA'
rJST(x, lexicon = NULL, S = 3, gamma = 5, ...)

Arguments

x

tokens object containing the texts. A coercion will be attempted if x is not a tokens.

...

not used

lexicon

a quanteda dictionary with positive and negative categories

K

the number of topics

S

the number of sentiments

alpha

the hyperparameter of topic-document distribution

gamma

the hyperparameter of sentiment-document distribution

beta

the hyperparameter of vocabulary distribution

alphaCycle

integer specifying the cycle size between two updates of the hyperparameter alpha

gammaCycle

integer specifying the cycle size between two updates of the hyperparameter alpha

Details

The rJST.LDA methods enable the transition from a previously estimated LDA model to a sentiment-aware rJST model. The function retains the previously estimated topics and randomly assigns sentiment to every word of the corpus. The new model will retain the iteration count of the initial LDA model.

Value

An S3 list containing the model parameter and the estimated mixture. This object corresponds to a Gibbs sampler estimator with zero iterations. The MCMC can be iterated using the fit() function.

• tokens is the tokens object used to create the model

• vocabulary contains the set of words of the corpus

• it tracks the number of Gibbs sampling iterations

• za is the list of topic assignment, aligned to the tokens object with padding removed

• logLikelihood returns the measured log-likelihood at each iteration, with a breakdown of the likelihood into hierarchical components as attribute

The topWords() function easily extract the most probables words of each topic/sentiment.

Author(s)

Olivier Delmarcelle

References

Lin, C. and He, Y. (2009). Joint sentiment/topic model for sentiment analysis. In Proceedings of the 18th ACM conference on Information and knowledge management, 375–384.

Lin, C., He, Y., Everson, R. and Ruger, S. (2012). Weakly Supervised Joint Sentiment-Topic Detection from Text. IEEE Transactions on Knowledge and Data Engineering, 24(6), 1134–-1145.

See Also

Fitting a model: fit(), extracting top words: topWords()

Other topic models: JST(), LDA(), sentopicmodel()

Examples

# simple rJST model
rJST(ECB_press_conferences_tokens)

# estimating a rJST model including lexicon
rjst <- rJST(ECB_press_conferences_tokens, lexicon = LoughranMcDonald)
rjst <- fit(rjst, 100)

# from an LDA model:
lda <- LDA(ECB_press_conferences_tokens)
lda <- fit(lda, 100)

# creating a rJST model out of it
rjst <- rJST(lda, lexicon = LoughranMcDonald)
# topic proportions remain identical
identical(lda$theta, rjst$theta)
# model should be iterated to estimate sentiment proportions
rjst <- fit(rjst, 100)