A Comparison between Factor Structure and Semantic Representation of Personality Test Items Using Latent Semantic Analysis

Abstract

In this thesis, semantic representation of personality test items is analyzed through LSA(Latent Semantic Analysis) and compared with factor structure to obtain implications for cognitive process of personality test item response. Generally, a survey response process consists of four stages: comprehension, retrieval, judgement and response. Traditionally, in order to investigate construct validity of the survey the final response data have been evaluated through exploratory factor analysis. But problem may occur when analyzing factor structure using response data only because such approach can overlook inconsistency of stimulus and response. In order to solve this problem, we need to take a direct approach to semantic representation of personality test items using LSA. In study 1, using passages generated in a limited context describing Big Five personality traits, a weighted document-term matrix is decomposed to produce semantic space. Furthermore, this thesis suggests semantic similarity matrix based on semantic space to compare factor structure with semantic similarity. Using this matrix, study 2 and 3 shows resemblance between factor structure and semantic similarity of personality test items. The results shows stimulus-response consistency. Moreover, the result implies that construct validity of personality test can be evaluated with its semantic representation without analyzing response data. Also, factor structure can be interpreted as semantic similarity of items.