Abstract: Person-job fit(PJF) is an essential task in talent recruitment. In the context of person-job matching, implicit feedback is more prevalent. Existing research mainly focuses on designing powerful interaction encoders to capture collaborative signals between jobseekers and positions, often overlooking the significance of loss functions and negative sampling. This paper centers on loss functions and negative sampling, and proposes a contrastive collaborative filtering method based on hard negative sample generation for person-job fit(HNCCF-PJF). This method refined the cosine contrastive loss function to address the issues of negative sample quality and bias. Specifically, it computed the cosine similarity between jobseekers and positions, setting a lower threshold to filter out easy negative samples and an upper threshold to mitigate the effects of false negative samples. Experiments on three benchmark datasets demonstrate that this approach outperforms latest benchmark methods in job-person matching performance.