Mining for attributes and values in tables

Table has been recognized as a simply and widely used data representation scheme. Each table alone typically contains rich and useful information which is valuable for many applications such as information retrieval, question-answering and etc. While all table formats can simply be parsed by human, this parsing is difficult for computer, prohibiting such applications to be done in an automatic manner. In this paper, we thus propose the comprehensive and novel table interpretation technique, namely tInterpreter. Essentially, it transforms a table into its corresponding horizontal 1-dimensional tables. To achieve this, the underlying work is based on (i) the similarity of two given cells with respect to the data type and the semantic correspondence concerns; (ii) the discovery for the boundary of a primitive table residing in a composite table; (iii) the identification of the attribute-value relationship and the value association of cells; and (iv) the integration of two pieces of similar or dissimilar information. The experimental result showed that the overall effectiveness of tInterpreter was higher than Chen, Tengli and Kim.