As the title indicates, this chapter addresses three goals. The first goal is to identify some important strengths of LSA, whereas the second is to identify some weaknesses. The third goal is to propose a few alternative quantitative models of representation with high-dimensional semantic spaces. As amply demonstrated by the range of applications described in other chapters of this book, there is no need to address the practical strength of LSA. Instead, we approach the first goal by presenting some basic facts about LSA in terms of simple algebra and demonstrate how powerfully this “data-mining” method captures human intuition. Limitations of LSA are observed from two perspectives: empirical evaluations of LSA in applications and formal analysis of LSAalgorithms and procedures. Regarding the third goal, we position LSAin a more general framework and then examine possible extensions. These extensions include three methods that adapt to learner perspective, context, and conversational history.
|Original language||English (US)|
|Title of host publication||Handbook of Latent Semantic Analysis|
|Publisher||Taylor and Francis|
|Number of pages||25|
|State||Published - Jan 1 2007|
ASJC Scopus subject areas