With the development of machine learning theory, increasing numbers of researchers would like to choose a CRF-based model to complete the same task. The traditional methods for opinion target extraction are based on frequency or manually designed rules, and it is not difficult to find obvious disadvantages, such as the fact that they are easily influenced by noise and have low flexibility, etc. Thus, we aimed to extract the opinion targets and opinion words from reviews. It can be seen that the research on “opinion target:opinion word” pair extraction is very meaningful. At the same time, e-commerce sellers can easily obtain consumers’ concerns and gain instant feedback to improve their service. In this condition, customers could be interested in different parts of the item, so they want to obtain the key information with its corresponding evaluation rather than a single sentiment polarity score to help them learn everything about the product. The sentence “The battery life seems to be very good, and have had no issues with it enabling the battery timer is useless.” includes two opinion targets, and each target matches one opinion word. Figure 1 shows an example of a sentence that has been condensed and whose opinion target and opinion word were retained. If these evaluation sentences are condensed and only the opinion target and opinion words are retained, it can help consumers to shop online, such as when booking hotels, ordering takeaways, etc., allowing users to quickly locate the information they need. With the popularity of mobile payment, the increasing numbers of consumers are causing the numbers of reviews of products to increase exponentially.
0 Comments
Leave a Reply. |