System and method for evaluating sentiment

摘要

An example system and method elicits reviews and opinions from users via an online system or a web crawl. Opinions on topics are processed in real time to determine orientation. Each topic is analyzed sentence by sentence to find a central tendency of user orientation toward a given topic. Automatic topic orientation is used to provide a common comparable rating value between reviewers and potentially other systems on similar topics. Facets of the topics are extracted via a submission/acquisition process to determine the key variables of interest for users.

主权项

1. A method for assessing sentiment of text, the method comprising using at least one programmed computer to perform steps including:
collecting a plurality of user reviews from a plurality of websites, each user review associated with a rating value of a plurality of rating values; identifying parts of text from the plurality of user reviews; determining a distribution of each part of text across the plurality of rating values; calculating, based on the distribution, a probability score representing a probability of a strength of each part of the parts of text for a corresponding rating value of the plurality of rating values; storing the parts of text, the distributions, and the probability scores in a training corpus; receiving a block of text associated with a new review; extracting one or more parts from the received block of text, where the parts are identified as a first type; accessing the training corpus to identify probability scores associated with the parts; performing a central tendency calculation for the new review by evaluating the probability scores associated with the parts; and assigning an orientation value to the new review based at least on the central tendency calculation, the orientation value reflecting sentiment of the received block of text, wherein the probability of the strength of a part of text for a particular rating value is degraded by the probabilities of the strength of that part of text for other rating values.