Given the analysis performed with real users, we can now return to our initial purpose in this paper and determine the effect of integrating an assistant within a critique-based recommender. Moreover, we will discuss whether any lesson can be learned and applied to the constructions of new approaches in the future.
In our experiments, we analyzed a critique-based recommender that only uses buttons, one that collects user’s preferences by means of an assistant, and a conversational recommender that integrates both the buttons and the assistant. Based on our results, which are shown in Fig. 5, we can conclude that: (1) the shortest ASL is obtained by integrating both the recommender system and the assistant, and (2) users perceive the integration of a cognitive assistant in a conversational recommendation framework as being of high quality and utility. The lesson learned is that users need to be involved in the recommendation process and they need to express their preferences in an easy-to-use interaction mechanism based on natural language, as shown in the usability analysis. Moreover, our conclusion is that a conversational recommender that is able to capture user needs expressed in natural language and involve her cognitively in the decision-making process is able to both guide the user appropriately in the search space and to learn their preferences faster.