Abstract: The traditional time series similarity search based on relevance feedback is to combine positive feedback with negative feedback to create new query vectors. This does not make full use of the value of negative feedback sequence, and it is easy to make too many changes to the initial query vector. This paper proposes a similar time series search method based on relevance feedback. The positive relevance feedback and negative relevance feedback are carried out separately. This way makes the results far away from negative relevant sequence. The results of similarity search on UCR data sets show that the similarity search method based on relevance feedback can improve the accuracy of similarity query.

Key words: data mining, time series, similarity search, relevance feedback