© 2019 IEEE. In recent years, the number of seniors living alone has increased, so the demand for the watching system has increased. In this paper, we propose a method to detect door-to-door sales from voice information with the aim of constructing a watching system to prevent consumer troubles of seniors living alone. In the proposed method, speech recognition is performed on conversation voice acquired from a microphone, and converted into text data. Subsequently, text data is input to the Doc2Vec model that has learned the conversation data on door-to-door sales and the cosine similarity is calculated. In this system, when the similarity exceeds a certain threshold value, it is determined that door-to-door sales is being conducted and a warning is issued to relatives. In order to evaluate the effectiveness of the proposed method, binary classification experiments of conversational text data were conducted. As a result, the area under the ROC curve drawn based on the cosine similarity was 0.94, suggesting that a valid decision was made.