As large language models (LLMs) continue to advance in intelligence, they are increasingly perceived as quasi-social agents, spurring the emergence of machine behavior studies centered on intelligent systems. A key issue in this domain is the extent to which LLMs exhibit human-like value alignments across various dimensions. Accurate identification of such tendencies is essential for developing a behavioral science of intelligent agents and fostering safer, more effective human-AI interaction. Existing literature has initiated exploratory inquiries from multiple perspectives, yet a systematic methodological framework remains lacking. This paper contributes to the academic discourse on value alignment in LLMs by proposing a typological framework based on the axes of “analytical scale” and “representational logic.” Through the analysis of four representative research approaches, the paper delineates core methodologies for examining value tendencies in LLMs. The findings aim to advance systematic understanding in this area and offer a novel methodological lens for the behavioral science of emerging LLM agents.