Systems, methods and devices for generating an adjective sentiment dictionary for social media sentiment analysis
Embodiments generally relate to systems and methods for generating a sentiment dictionary and calculating sentiment scores of adjectives within the sentiment dictionary. A set of seed words can be identified and expanded using synonyms and antonyms of the set of seed words. Social media data can be parse to identify adjectives that link to the set of seed words with the words “and” or “but.” Matrices representing the attraction and repulsion among the linked adjectives can be generated. A factorization algorithm can be minimized to determine an output matrix that comprises positive and negative sentiment scores for each of the adjectives. In embodiments, a sentiment score for part of all of the social media data can be calculated using the output matrix, and one or more parts of the social media data can be classified as a positive or negative sentiment.
1. A method of processing data, the method comprising:
identifying a set of seed words comprising words defined as either positive words or negative words;
extracting, from a set of data, adjectives linked to the set of seed words with “and”;
extracting, from the set of data, adjectives linked to the set of seed words with “but”;
determining a first value indicating a first frequency with which the adjectives are linked to the set of seed words with “and”;
determining a second value indicating a second frequency with which the adjectives are linked to the set of seed words with “but”;
calculating a synonym score using an equation x ij =w ij + +Log(d ij + +1), wherein x ij corresponds to the synonym score, w ij + corresponds to a number of times that a word i and a word j appear to be synonyms, and d ij + corresponds to the first equency; and
calculating, by a processor, sentiment scores for each adjective of the set of data based on the synonym score and an antonym score calculated using the second value.
2. The method of claim 1 , wherein identifying the set of seed words comprises:
identifying the set of seed words; and
generating an expanded set of seed words by adding synonyms and antonyms of the set of seed words to the set of seed words.
3. The method of claim 1 , wherein the set of seed words comprises a positive set and a negative set.
4. The method of claim 1 , wherein each of the sentiment scores comprises a positive sentiment score and a negative sentiment score.
5. The method of claim 1 , further comprising calculating the antonym score using an equation c ij =I ij +Log(d ij − +1), wherein c ij corresponds to the antonym score, I ij equals 1 if the word i is in a positive set of the set of seed words and the word j is in a negative set of the set of seed words, or vice versa, and equals 0 otherwise, and d ij − corresponds to the second frequency of.
6. The method of claim 5 , wherein the sentiment scores can be calculated by minimizing an equation ∥X−HSH T ∥ F 2 +αTr(H T CH), wherein:
X is a nonnegative symmetric graph matrix formulated using the synonym score;
H is a matrix comprising the sentiment scores;
S is a modifier to allow H to be closer to a form of cluster indicators;
α is an input parameter;
C is a constraint weight matrix formulated using the antonym score; and
∥ ∥ F indicates a Frobenius norm.
7. The method of claim 1 , further comprising:
classifying each adjective of the set of data as either a positive word or a negative word.
8. The method of claim 1 , wherein the set of data is social media data.
9. The method of claim 1 , further comprising:
determining a sentiment score for at least part of the set of data.
10. A system for processing data, comprising:
an interface to a storage device configured to store a set of data; and
a processor, communicating with the storage device via the interface, the processor being configured to:
identify a set of seed words comprising words defined as either positive words or negative words;
extract, from the set of data, adjectives linked to the set of seed words with “and”;
extract, from the set of data, adjectives linked to the set of seed words with “but”;
determine a first value indicating a first frequency with which the adjectives are linked to the set of seed words with “and”;
determine a second value indicating a second frequency with which the adjectives are linked to the set of seed words with “but”;
calculate a synonym score using an equation x ij =w ij + +Log(d ij + +1), wherein x ij corresponds to the synonym score, w ij + corresponds to a number of times that a word i and a word j appear to be synonyms, and d ij + corresponds to the first frequency; and
calculate, by a processor, sentiment scores for each adjective of the set of data based on the synonym score and an antonym score calculated using the second value.
11. The system of claim 10 , wherein identifying the set of seed words comprises:
identifying the set of seed words; and
generating an expanded set of seed words by adding synonyms and antonyms of the set of seed words to the set of seed words.
12. The system of claim 10 , wherein the set of seed words comprises a positive set and a negative set.
13. The system of claim 10 , wherein each of the sentiment scores comprises a positive sentiment score and a negative sentiment score.
14. The system of claim 10 , wherein the processor is further configured to calculate the antonym score using an equation c ij =I ij +Log(d ij − +1), wherein c ij corresponds to the antonym score, I ij equals 1 if the word i is in a positive set of the set of seed words and the word j is in a negative set of the set of seed words, or vice versa, and equals 0 otherwise, and d ij − corresponds to the second frequency.
15. The system of claim 14 , wherein the sentiment scores can be calculated by minimizing an equation ∥X−HSH T ∥ F 2 +αTr(H T CH), wherein:
X is a nonnegative symmetric graph matrix formulated using the synonym score;
H is a matrix comprising the sentiment scores;
S is a modifier to allow H to be closer to a form of cluster indicators;
α is an input parameter;
C is a constraint weight matrix formulated using the antonym score; and
∥ ∥ F indicates a Frobenius norm.
16. The system of claim 10 , wherein the processor is further configured to:
classify each adjective of the set of data as either a positive word or a negative word.
17. The system of claim 10 , wherein the set of data is social media data.
18. The system of claim 10 , wherein the processor is further configured to:
determine a sentiment score for at least part of the set of data.
19. The method of claim 1 , wherein:
the sentiment scores are calculated using graph matrixes; and
the synonym score and the antonym score indicate graph edge weights.
20. The system of claim 10 , wherein:
the sentiment scores are calculated using graph matrixes; and
the synonym score and the antonym score indicate graph edge weights.