Automatic Tag Recommendation Approach with Keyphrase Extraction and Word Embedding Techniques

Taechawat Konkaew ； Sukumal Kitisin

《電腦學刊》 30卷2期 (2019/04) Pp. 135-149

https://doi.org/10.3966/199115992019043002012

摘要

The number of online videos made available has been increasing rapidly. The most common way to find interesting videos is to search for the relevant tags. However, a large number of video clips may not have tags which cover in every aspect their contents, therefore searching for the relevant videos may not be very effective. Recent development in topic modeling research shows good performance of the word embedding model [1-2]. Our work proposed an automatic video tag recommendation approach from video transcript using a hybrid of unsupervised keyphrase extraction and word embedding model considering the semantic similarity between keyphrase candidates. Our experiments were performed on TED Talk videos dataset. The results show improvements over a variety existing approach. The proposed approach can help video owners identify a set of tags presumably having good coverage of the video contents. Our work could be applied in many existing video sharing platforms like YouTube, Twitch, etc.

關鍵字

automatic tag recommendation ； keyphrase extraction ； natural language processing ； word embedding