Context Aware Recommendation System

Abstract

Life is becoming mechanical with passage of time and people want to spend the least amount of time to find the best fit things or commodities for them. Who doesn’t need a personalized agent that can help them find perfect matches for their requirements; quickly and efficiently? Considering this need into account, several recommender systems have emerged into market mostly as website functionalities, but a very few as chatbots. Problems with those recommender systems are that one, they are not in the form of a natural language, user friendly chatting platform hence customers can’t interact with them easily and systems tend to seek hidden meanings from their conversation which results in catastrophic misunderstanding. And even if some chatbots are designed for human like conversation, they cannot maintain context, store implicit user preferences, click through data, and hence are unable to perform according to multiple types of data.

The solution to the problems stated above is a bot that is contextually aware and recommends the users what they want. The domain of our project is real estate as this project is a collaboration with the real estate website Aarz.pk. We designed an assistant in the form of a chatbot that has been deployed on Aarz.pk website. This Aarz bot is designed to handle user queries in two different languages: English and Urdu. Moreover, since the real estate agents in Pakistan usually use a mix of English and Urdu while chatting, we have developed a transitional module to cater Roman Urdu as well. Now the users can have an enhanced experience while chatting to this bot as it will store one's preferences for later use. This means that if suppose one is new in town, looking for a place to live in they can easily be assisted by the Aarz bot which will not only use the requirements one entered but will also use the customers activities like the clicks on the website to display the most relevant property on top. Hence, Aarz bot can handle queries of the users in such a way that it establishes context and checks if the present query has any relation with previous queries or not and generates response based on that. Those responses are then sorted according to relevance based on user history and their likes and dislikes. This approach helps in creating uniquely satisfying experience for the users. We have implemented our project using combination of both traditional algorithms and machine learning approach. Our system has four major components. It contains a chatbot, database of user history and property data, recommendations generation and ranking of recommendations in most relevant order. Chatbot converses with the user, recognizes and gets information from the user then store the information in a database. The main technique used in this project revolves around Deep Ranking.