| |
|
|
| |
| |
A leading global academic institution specializing in text analysis. Focus has been in providing solutions to understanding and analyzing the sentiment of texts in product review sites.
|
|
| |
| |
| |
Sentiment analysis is concerned with extracting opinionated content and analyzing it so as to provide some clue how users feel about some issues/things. These opinions which are often present in user generated contents are valuable information for companies to determine brand or product reputation, and to identify the faults in their products. General-purpose sentiment lexicons have been proposed, for example SentiWordNet, Subjectivity Lexicon and General Inquirer. However, since the sentiment of words differ across different domains, general-purpose sentiment lexicons are ineffective in providing accurate sentiment analysis as each word in the lexicons is attached with a fixed polarity value and may not be representative of the sentiment. Furthermore, the sentiment of words within the same domain might differ depending on the context. For example, consider the mp3 player domain, the word “tiny” is positive for the size product feature or aspect but negative for the memory aspect. To the best of our knowledge, there is no such context and domain sensitive sentiment lexicon currently available. As such, we have developed a method to generate a context and domain sensitive sentiment lexicon which will be useful in sentiment analysis tasks in the product domain. The lexicon consists of entries from different product domains, and for each domain, a list of aspects and their associated sentiment words are generated. The lexicon assigns a polarity, either positive or negative to each sentiment word.
Our method is an automated one; this contrasts with a manual generation method which is tedious and is not scalable over different domains. It includes an aspect/sentiment extractor that can extract aspects and their associated sentiment words from product reviews, which when combined a context and domain sensitive lexicon can be generated. This process is fully automated with no human intervention.
The aspect/sentiment extractor allows users to extract the different product features or aspects of a product and to determine the sentiment for each feature. For example, given some review text, “Battery life is shocking, have to fully charge every day, its down to zero by nite”, our system can extract Battery Life as an aspect and shocking as the associated negative sentiment for this aspect.
Since the lexicon considers the domain and its aspects, the context and domain sensitive sentiment lexicon is able to provide a more accurate analysis of sentiment words in a product review compared to other general purpose sentiment lexicon. This lexicon can be used in a sentiment analyzer system which together can benefit both product companies and consumers. Companies can now more accurately analyze the opinions about their own products or their competitors’ products and have an automated way to process the opinions without having to read through the many reviews about their products. Specifically, such a system allows product-producing companies to discover problems with their products and allows them to better understand the needs of their users. On the other hand, consumers will find the system effective in assisting them to understand the reviews of the products so that they can make wiser decision on which products fit their needs.
|
|
| |
| |
| |
Looking for partners to collaborate and develop commercial application on top of the technology.
|
|
| |
|