This study introduces a novel approach for automated hypothesis generation in psychology by integrating large language models (LLMs) with causal graphs. By analyzing 43,312 psychology articles using an LLM, the researchers extracted causal relationship pairs and constructed a specialized causal graph. Applying link prediction algorithms, they generated 130 potential psychological hypotheses focusing on "well-being." These hypotheses were compared against those generated by doctoral scholars and LLMs alone. The combined approach of LLM and causal graphs outperformed LLM-only hypotheses in terms of novelty, as evidenced by statistical significance (t(59)=3.34, p=0.007 and t(59)=4.32, p<0.001). Deep semantic analysis further confirmed this alignment. The study's methodological framework involves three steps: literature retrieval, causal pair extraction, and hypothesis generation. The literature retrieval phase involved downloading ~140,000 psychology-related articles from public databases. Causal pair extraction used GPT-4 to identify causal relationships, resulting in a causal relationship network based on 43,312 selected articles. Hypothesis generation utilized link prediction algorithms to forecast potential causal relationships. The study's results show that combining LLMs with causal graphs can revolutionize automated discovery in psychology, extracting novel insights from extensive literature. The study also provides novel tools and methodologies for causal analysis and scientific knowledge discovery, representing the fusion of modern AI with traditional research methodologies. This integration bridges conventional theory-driven methodologies in psychology with data-centric research paradigms, enriching our understanding of factors influencing psychology, especially in social psychology. The study's findings highlight the effectiveness of the LLM-based causal graph (LLMCG) framework in generating hypotheses with high novelty and usefulness. The LLMCG framework outperformed both LLM-only and human-generated hypotheses in terms of novelty, as demonstrated by statistical analysis. The study also showed that the LLMCG framework produces hypotheses with deeper conceptual incorporations and a broader semantic spectrum. The results underscore the potential of integrating LLMs with causal graphs for hypothesis generation in psychology.This study introduces a novel approach for automated hypothesis generation in psychology by integrating large language models (LLMs) with causal graphs. By analyzing 43,312 psychology articles using an LLM, the researchers extracted causal relationship pairs and constructed a specialized causal graph. Applying link prediction algorithms, they generated 130 potential psychological hypotheses focusing on "well-being." These hypotheses were compared against those generated by doctoral scholars and LLMs alone. The combined approach of LLM and causal graphs outperformed LLM-only hypotheses in terms of novelty, as evidenced by statistical significance (t(59)=3.34, p=0.007 and t(59)=4.32, p<0.001). Deep semantic analysis further confirmed this alignment. The study's methodological framework involves three steps: literature retrieval, causal pair extraction, and hypothesis generation. The literature retrieval phase involved downloading ~140,000 psychology-related articles from public databases. Causal pair extraction used GPT-4 to identify causal relationships, resulting in a causal relationship network based on 43,312 selected articles. Hypothesis generation utilized link prediction algorithms to forecast potential causal relationships. The study's results show that combining LLMs with causal graphs can revolutionize automated discovery in psychology, extracting novel insights from extensive literature. The study also provides novel tools and methodologies for causal analysis and scientific knowledge discovery, representing the fusion of modern AI with traditional research methodologies. This integration bridges conventional theory-driven methodologies in psychology with data-centric research paradigms, enriching our understanding of factors influencing psychology, especially in social psychology. The study's findings highlight the effectiveness of the LLM-based causal graph (LLMCG) framework in generating hypotheses with high novelty and usefulness. The LLMCG framework outperformed both LLM-only and human-generated hypotheses in terms of novelty, as demonstrated by statistical analysis. The study also showed that the LLMCG framework produces hypotheses with deeper conceptual incorporations and a broader semantic spectrum. The results underscore the potential of integrating LLMs with causal graphs for hypothesis generation in psychology.