Enhanced Twofold-LDA Model for Aspect Discovery and Sentiment Classification

Enhanced Twofold-Lda Mode

Figure 1 shows an example of the standard LDA model output, usually 30 to 100 topics manually labelled with the aspects that relate to the words in each topic. This manual work of making sense of the output can be quite time-consuming and would not be easily understood by an end user.

Figure 1.

Standard LDA output (Blei, Ng & Jordan, 2003)

Previous research using natural language processing techniques shows information in graphical form which is much more user friendly. Figure 2 shows an example of a chart comparing two digital cameras. From the chart we can clearly see that Digital Camera 1 is better in nearly every aspect and both cameras have equal positive and negative opinion on their weight. This chart means that customers no longer need to read through countless reviews to find the useful information they require. Also, the chart allows a customer to compare 2 products alongside each other.

Figure 2.

Visual comparison of two digital cameras (Liu, Hu & Cheng, 2005)