Development of Domain-Specific Lexicon for Aspect-Based Sentiment Analysis

Aspect term extraction is one of the main subtasks in aspect-based sentiment analysis. An aspect extraction method based on the Sequential Covering algorithm successfully used a list of aspects and opinion words to improve the performance of aspect extraction. The aspect and opinion list used in the method is crafted manually, costing a significant amount of time and effort. To ease the effort, we proposed a way to automatically build the aspect and opinion list. We also modified the scope of the word list to a larger domain. The resulting word list is therefore called domain-specific lexicon. In this paper, we used word embedding technique to develop domain-specific lexicons of size 1000. The data used in the development of domain-specific lexicon were collected from review websites using a focused web crawler. The resulting domain-specific lexicons were used in the modified Sequential Covering method. We used a total of 3,124 review sentences from the digital camera, handphone, and restaurant domain to evaluate the performance of the modified Sequential Covering method. Experimental results showed that this method managed to yield better F1 scores than the baseline Sequential Covering method. The best F1 scores for each dataset were as follows: 0.645 for Nikon Coolpix 4300, 0.581 for Canon G3, 0.629 for Nokia 6610, and 0.705 for ABSA16_Restaurants_Train_SB1.