Article Preview
Top1. Introduction
With the rapid increasing of the volume and scale of Web data, Web has been one of the major sources to acquire competitive intelligence for enterprises. Enterprise competitive intelligence refers to the intelligence related with competitors, competitive environment, and competitive strategies (Zhao and Jin, 2009), among which the competitive strategies is the most difficult type of competitive intelligence to obtain in real applications. The competitive strategies about a certain company usually hide in a lot of phenomenon. For example, the future developing strategy of a company, e.g., IBM, may be reflected by its acquirement actions. There is a real case regarding IBM. Recently, this company is advancing its development on big data analysis (IBM, 2015). But if we check the acquirement history of IBM in recent years, we can find that since 2009 it keeps buying companies on data analysis, such as SPSS, Emptoris, and Netzza. Thus, if we can extract the company acquirement relation about concerned competitors from the Web and detect their early intends in strategical development, it will be helpful to conduct deeply analysis on enterprise competitive intelligence (Zhao and Jin, 2010a).
Company acquirement relation can be represented as a triple <ORG1, verb_tense, ORG2>, where ORG1 is the acquirer while ORG2 is the acquiree and verb_tense refers to the tense of the acquiring action. The verb_tense usually indicates the state of the acquirement relation. For instance, in a sentence saying “IBM has acquired Algorithmics”, we can detect that there exists an acquirement relation between IBM and Algorithmics and this action has been actually completed due to the verb tense “has acquired”.
Compared with traditional techniques in relation extraction, the extraction of company acquirement relation in the Web has two key issues:
In English sentences, the sentence tense has an important impact on the effectiveness of company acquirement relations extraction. For example, the sentence “IBM has acquired Algorithmics” is with the tense of present perfect, which indicates the validness of the acquirement relation between IBM and Algorithmics. However, in a sentence with future tense such as “IBM will acquire Algorithmics”, the acquirement relation may not be valid. Another challenging issue is that some sentences imply negative semantics, e.g. “IBM rumored to acquire Algorithmics”.
Web information usually has multiple styles, as different Web information sources have differing format when describing events or news. Some typical sources of Web information are blog, bulletin board system (BBS), Wikipedia, and news pages. Besides, Web pages may differ in the time factor, e.g., the publication dates of Web pages are usually different. As a result, there are various forms of company acquirement relations in the Web, so we cannot simply detect them by analyzing one Web page. The more reasonable way is to check a set of related Web pages and detect their semantic strength in asserting the validness of a possible company acquirement relation. This problem is essentially an issue of classification, as we have to find the Web pages showing positive semantics and others showing negative semantics, and then make further decision on the validness of company acquirement relations.
Traditional ways in relation extraction only considered the limited characters of sentences, such as morphology and syntax (Zhou et al., 2005; Zhang et al., 2005; García and Gamallo, 2011), and lack of the analysis on the tense of sentences as well as the semantic strength of sentences in determining the validness of relations. Some recent works showed that the tense of sentences has a very important impact on relation extraction (Zhao and Jin, 2010b). Therefore, we introduce the tense of sentences into the extraction of company acquirement relations.