A Method of Refining Evaluation Criteria According to Purpose of Use Based on Consensus of Distributed Representations

This research proposes a new method for subdividing evaluation items based on customer review data according to the intended use of the evaluation target. It is important to set the five-point evaluation items appropriately, and we point out that the hierarchical structure of evaluation items varies significantly depending on the intended use of the evaluation target. By applying an ensemble of word embeddings to corpora constructed for each intended use, we were able to achieve a clearly distinct subdivision of evaluation items for each intended use. By applying this to Rakuten Travel review data, we confirmed that it is possible to automatically extract candidate evaluation items according to the intended use of accommodations from actual data. The proposed method is expected to further automate the subdivision of evaluation items.