Ding and Liu, SIGIR 2007

From ScribbleWiki: Analysis of Social Media

Jump to: navigation, search


The Utility of Linguistic Rules in Opinion Mining


Can be found here.


This paper tries to estimate the significance of using linguistic rules for opinion mining. Their primary contribution lie in two directions. They first propose a function for calculating the semantic orientation of a particular feature as expressed in a customer review. They also investigate the use of some linguistic rules that help finding out the semantic orientation of an opinion word. In previous works on opinion mining, researchers mostly have tried to conclude about a review by looking at particular opinion words like "good", "bad", "amazing" etc. These words have a particular semantic orientation and may help in analyzing a review, but the authors claim that there are various other words whose semantic orientation depends on their context.

As a first contribution, the authors propose a function that helps in finding out the opinion about a particular feature of a product. The authors first segment each sentence into segments that correspond to different features just by using BUT word/phrases. For each segment S_k and a feature f_i, the score is nothing but the summation of the semantic orientation of each opinion word in the segment divided by the distance of the feature and the opinion word. The divisor, i.e. the distance value penalizes the opinion words that are far away from the feature. The semantic orientation of an opinion word is 1 for a positive opinion and -1 for negative.

Next, the authors propose the use of linguistic rules that help find the opinion of a particular word. The rules are:

  • Intra-sentence conjunction rule: To find out the semantic orientation of a particular word, the algorithm tries to find out related reviews where the word has appeared, and whether it appeared with an already known positive or negative opinion word in a conjunctive form.

  • Pseudo intra-sentence conjunction rule: This rule is similar to above, but it uses reviews where the concerned opinion word appeared, but not in an explicit conjunctive form. For example, if the given segment is "the battery life is long" and we are trying to find out the orientation of "long", there can be another review where we find a segment "the camera has a long battery life, which is great". Here, long appears with "great" which is a known positive orientation word, but not in an explicit conjunctive form.

  • Inter-sentence conjunction rule: This rule tries to extend the above relationships to consecutive sentences.

The authors also use a synonym-antonym rule, when synonyms' or antonyms' semantic orientations are concluded from a known word.

Experiments were conducted using 742 customer reviews from 14 products from amazon.com. It was observed that the linguistic rules improved the score in comparison to a baseline system. Results were also compared with FBS and OPINE systems and it showed improvements by a good extent.

Personal tools
  • Log in / create account