<p>The primary objective of knowledge retrieval is to address conversational queries. Recent studies have simplified conversational search by highlighting response ranking and answering questions within a conversation, where the response is retrieved from a candidate set or extracted from a passage. With the rise of conversational AI systems such as Amazon Alexa, Google Assistant, and Apple Siri, multi-turn question answering has garnered significant research attention. In conversational search, the answer may not directly stem from the current question but could be linked to preceding questions within the ongoing conversation. Therefore, considering the history of previous questions plays a crucial role in formulating responses to current questions. This paper introduces a straightforward yet effective method for conversational machine comprehension (CMC) using a fine-tuning approach with a BERT encoder. The proposed model incorporates a key history selection module, emphasizing the selection of pertinent conversation history and generating a history context vector of interest. The integrated model, with its history selection module and multi-layered architecture, exhibits improved performance compared to existing models. Experimental results demonstrate that the proposed model attained an overall F1-score of 68.0, outperforming models such as BERT + HAE (63.1) and HAM (66.7).</p>