With the advancement of service-oriented computing, many online services have provided convenience to users. Operators are required to analyze the large amounts of log data generated by the system to maintain quality of service (QoS). Log data is characterized by its large volume, semi-structured format, and dynamic, making manual analysis prohibitively costly. Researchers have employed deep learning-based methods for automated log analysis, which has shown some progress, but issues regarding high-labeled costs, interpretability, and usability persist. In this paper, we consider a more practical scenario involving a small amount of abnormal logs, a large volume of unlabeled logs, and some degree of anomaly contamination. To this end, we design a weakly supervised model using spike neural networks to detect system anomalies. Moreover, we co-designed a log question-answering model, which enables operators to analyze log data through natural language for efficient anomaly handling. Experimental results demonstrate the effectiveness of our method across both anomaly detection and question-answering tasks.