Product Feature Extraction and Sentiment Analysis in Product Reviews
Team Members
Project Title
Product Feature Extraction and Sentiment Analysis in Product Reviews
Project Abstract
In this project, we plan to work on product reviews of various product classes and analyze them for finding the product features and opinion of various customers about those features. Using this analysis we aim to identify feature-wise good and bad aspects of a given product. This can be a useful practical solution to allow customers to help decide how well a product satisfies his/her needs if they are only looking for few important features in a product and don't care about other features.
Some of the product classes do not have well-defined features like movies, books etc. For such classes, we need to identify implicit features based on what customers liked/disliked about it, which could be something like the movie plot, specific actor's performance etc. Also, we aim to analyze the performance of different feature extraction techniques on different product classes. Opinion mining for specific product features would require some level of semantic understanding to separate opinions about other features mentioned in the same review. Also same word can be used to express contrasting opinions, which must be taken into account to avoid incorrect sentiment classification if only a global polarity is used for each sentiment word.
Based on these challenges, we aim to achieve a robust solution for extracting features and opinion about them from the product reviews.
Datasets
Amazon product reviews data set -
- http://www.cs.uic.edu/~liub/FBS/sentiment-analysis.html#datasets
- Amazon product review dataset for various classes
Apart from that, we have already built our own web crawlers and extracted some more product reviews from www.amazon.com to train our system.
Baseline
We suggest the following baselines -
- Product Feature Extraction: We can use an n-gram model to extract noun phrases/words which are usually candidate features for a product. We can also add Wordnet sysnset data to expand on the list of candidate features and also put an appropriate frequency threshold to discard unimportant features.
- Sentiment Analysis: We can use a list of sentiment words already marked as positive and negative and then score each sentence as positive or negative or neutral based on presence of these words in that sentence.
Related Papers
- Pang, B., L. Lee, and S. Vaithyanathan. 2002. Thumbs up?: sentiment classification using machine learning techniques. In Proceedings of the ACL-02 conference on Empirical methods in natural language processing-Volume 10, 79–86.
- Turney, P. D. 2002. Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, 417–424.
- M. Hu and B. Liu. Mining Opinion Features in Customer Reviews. In Proceedings of Nineteenth National Conference on Artificial Intelligence. 2004.
- Dave, K., Lawrence, S., and Pennock, D.M. 2003. Mining the peanut gallery: Opinion extraction and semantic classification of product reviews. WWW 2003.
- Jianxing Yu , Zheng-Jun Zha, Meng Wang, Kai Wang, Tat-Seng Chua, Domain-Assisted Product Aspect Hierarchy Generation: Towards Hierarchical Organization of Unstructured Consumer Reviews, EMNLP 2011.
- Bishan Yang and Claire Cardie, Extracting Opinion Expressions with semi-Markov Conditional Random Fields.
