The majority of current approaches attempt to detect the overall polarity of a sentence, paragraph, or text span, regardless of the entities mentioned (e.g., laptops, restaurants) and their aspects (e.g., battery, screen; food, service)
Our task was concerned with aspect based sentiment analysis (ABSA), where the goal was to identify the aspects of given target entities and the sentiment expressed towards each aspect.
Github code: https://github.com/AkshitaJha/IRE
Project Web Page: juhi-ghosh.github.io/IRE
Youtube: https://youtu.be/ksfcodFeVHg
3. Sentiment Analysis
❖ Use of NLP, text analysis and linguistics to identify and extract
subjective information in sentence.
❖ Used to identify attitude of speaker with respect to some topic.
❖ Two types of sentiment analysis
➢ Subjectivity/Objectivity Identification
➢ Feature/Aspect Based Sentiment Analysis
4. Problem Statement
❖ Given a product review containing multiple features and varied opinions.Extract
➢ Aspect term
➢ Aspect term polarity
➢ Aspect Category
➢ Aspect Category Polarity
❖ Datasets Used -
➢ Semeval 2014 dataset for restaurant reviews.
➢ Corpus developed by Cruz-Garcia for category extraction.
5. Subtask 1: Aspect term extraction
Given a set of sentences with pre-identified entities (e.g., restaurants), identify the
aspect terms present in the sentence. Eg:
“I liked the service and the staff, but not the food”
Aspect Terms: servıce, staff, food
6. Subtask 2: Aspect Polarity
For a given set of aspect terms within a sentence, determine whether the polarity of
each aspect term is positive, negative or neutral. Eg.
“I liked the service and the staff, but not the food”
Aspect Polarity: {servıce, posıtıve}
{staff, posıtıve}
{food, negatıve}
7. Subtask 3: Aspect Category Detection
Given a predefined set of aspect categories (e.g., price, food), identify the aspect
categories discussed in a given sentence.
Categories: {food, service, price, ambience, anecdotes/miscellaneous}
“I liked the service and the staff, but not the food”
Category Detection: {servıce, food}
8. Subtask 4: Aspect Category Polarity
Given a set of pre-identified aspect categories (e.g., {food, price}), determine the
polarity. Eg:
“I liked the service and the staff, but not the food”
Aspect Category Polarity: {servıce, posıtıve}
{food, negatıve}
9. References
❖ Subhabrata Mukherjee, Pushpak Bhattacharyya “Feature Specific Sentiment
Analysis for Product Reviews”
❖ Soujanya Poria, Erik Cambria “A Rule-Based Approach to Aspect Extraction
from Product Reviews”
10. Tools Used
❖ POS-Tagger - Get pos tags for the sentence.
❖ Stanford Parser - Extract relationships among different words of a sentence.
❖ TextBlob - Identify the polarity of the aspect terms
❖ Wordnet - A lexical database for the English language
11. Approach
❖ Developed rules to identify aspect terms in a given restaurant review using
relationships obtained from stanford parser.
➢ advmod, agent, amod, dobj, nsubj, pobj, xcomp
➢ all nouns and adjectıves
➢ search upto a depth of 2 ın Dependency Relatıons
❖ Identified polarity of each aspect term using TextBlob.
❖ Created a dictionary of aspect categories using corpus developed by Cruz-
Garcia. Also, add synonyms and antonyms.
❖ Mapped each aspect term to the corresponding categories.
16. Challenges
❖ Identıfyıng aspects - Getting accurate aspects itself is a big challenge. Nouns
may not always be aspect terms.
❖ Aspect term polarıty - Anaphora resolution poses a problem in identifying
aspect term polarity.
❖ Aspect based sentıment - Sarcasm and contrastive conjunctions alter the
meaning of entire sentence. Crucial in determining sentiment.
❖ Errors in POS Tagger and Stanford Dependency Parser
17. Conclusions
❖ Proposed framework leverages on common sense knowledge and on dependency
structure and thus, is unsupervised.
❖ The work exploits associations between the opinion expressions about various
features that form a coherent review using dependency parsing.
❖ In-depth analysis of dependency relations does not seem significant.
❖ We do not detect sarcasm and humour in our system.
18. Future Work
❖ Discover more rules for aspect term extraction.
❖ Combine existing rules for complex aspect extraction.
❖ To extract aspect categories dictionaries created can be made more noise free.
❖ Detect conflict sentiment
❖ Hybrid approach or supervised classification can be used for better
performance.
19. Other resources....
Youtube Video: https://youtu.be/ksfcodFeVHg
Project Web Page: juhi-ghosh.github.io/IRE
Github Repository: https://github.com/AkshitaJha/IRE