Mining Similar Traces of Entities on Web

Abstract

Events about entities have been widely collected on Web, allowing us to analyze how peer entities interact and learn the relationships that exist among the entities. In this paper we investigate similar traces that have not been adequately studied so far. Intuitively, peer entities tend to have similar traces. The challenges in mining similar traces are: (1) the occurring time lags of traces are usually unknown and varying; (2) the existence of large-scale events of entities and complexity of the model representing all the events. In this paper we propose a simple, but practical method that addresses all these challenges. Firstly, sliding windows are adopted to filter out the significant events and then find the candidate topic sequences. Secondly, dynamic programming is employed to mine similar candidate topic sequences of entities. Finally, an efficient method is proposed to mine all the similar traces of entities. It is able to mine similar traces of peer entities with high accuracy. We conduct comprehensive experiments on synthetic datasets to demonstrate the efficiency of the method proposed.

Addendum to “Mining Similar Traces of Entities on Web” (published in Vol. 15, No 6)

“…

The work presented in Section 5 of [1] differs from the work in Section 2 of [2] due to the following reasons.1. Different research objects and different meaning of a sliding time window.

…”

“…The work presented in Section 5 of [1] differs from the work in Section 2 of [2] due to the following reasons.…”

“…Different research objects and different meaning of a sliding time window. Our work presented in [1] is based on massive events data sets in a long history which have already been extracted, while the work presented in [2] is based on the context in real-time from the live public data stream of Twitter. So the scale of data to process in the case of [2] isn't in the same order of magnitude as ours, where the same approach has a different meaning for our work.…”

“…In order to further distinguish between the two works mentioned above, Section 5 of [1] is extended with more details below.…”

“…Our work presented in [1] is based on events data sets in a long history which have been extracted already. If the target entity is TE, the pre-processing work before significant events identification is done through a separate task to aggregate all the events of TE, then count the value of idf(e) of each event e and finally construct the event relationship graph of TE in which all the edges are based on the co-occurrence of events.…”