Automatic Ranking of Product Reviews According to Helpfulness (Yissum)
A system for sorting through thousands of book reviews to rank the most helpful
Identifies helpful information in reviews and automatically ranks them
Human evaluators’ choice of most helpful reviews corresponded with the
MLLM’s choice 85% of the time
Can be used for all types of product reviews, such as consumer electronics
Overcomes biases found in user voting mechanismsA Multi Layer Lexical Model (MLLM)-based algorithm for ranking book reviews. The MMLM approach is a system for data mining and content analysis that examines book reviews in order to establish which of the reviews are the most helpful. If available, the text of the book itself can also be used to enhance the output. The layers contain compact, high-quality lexicons of words specific for each layer, such as terms common in product reviews, specific lexical terms connected with the type of book and terms connected with the title.System outperforms voter ranking and random sampling
System provides a continuous scale of grading
Allows helpful reviews that may potentially be overlooked to be identified
Can easily adapt the review ranking to match different criteria, such as review length
Fully unsupervised approach precludes the need for human annotations. does not depend on active users– reduces costsSystem was developed using books that had large numbers of reviews. Future development will be for a system that works where there are fewer reviews
The MLLM approach will be used to generate a single comprehensive review from the reviews ranked most helpful
Can be applied to reviews of all sorts of products to assist consumers make purchasing decisions
PCT 09087636
10-2008-2072
以色列
