Topic Title: Decision Support in Purchase Process
Technical Area: Recommender System, knowledge graph, fuzzy cognitive maps
Purchase is a decision-making process. The action (interaction) sequence of users may reflect the decision making in the purchase. In the current practice, such knowledge is mined through the data of user behavior or action logs. However, it is not always the case that the action sequence reflected the decision-making process. A user may leave the most important but obvious options last, while concentrating on the indecisive options; another user may start with the most important factors and then the rest.
Further, data-based mining from the log of user behaviors is often inefficient and sometimes ineffective to understand the decision-making process. It takes a long time to accumulate sufficient history data about a user to fully understand the user’s habit and decision model. A user’s purchase decision can be based on different reasons. For example, he may choose to purchase a 400L refrigerator simply because the size fits in his design (not considering the price factor), or it is the mode with largest volume yet still within his budget range (very sensitive to price), or it has a special panel design which suits his taste (willing to pay high price for quality product or special designs). Such decision knowledge is not contained in the interaction sequence of the purchase process, making it difficult and inefficient to learn such decision knowledge from history data. It requires a long history and a large volume of history data to mine the users purchase pattern. Additionally, when it involves different categories of products or influence by “expert” views, the behavior can be even more difficult to be mined.
This project will develop a new knowledge graph model to support users’ purchase process, acquire and analyze the causal relationship knowledge underlying the purchase decision, which can be further applied to support the user and other users’ purchase decision. A ‘reasonable’ interaction model will be developed to capture the decision knowledge and process of users, integrate with experts’ views, and better support users as well as performing precise recommendation.
Related Research Topics
- Firstly, we will select one (or a few) product that involves reasonably complex decision making. Such a product normally has many aspects affect users’ purchase decision. Some aspects could lead to extra difficulties in data mining. Purchase frequency one of them. For example, a refrigerator has many aspects contributing to the decision process, including volume, size, doors, brands, engine type, and etc. Users do not purchase refrigerators often. Therefore, there are no much history data that can be relied on. The last purchase could be 3 to 5 years ago, when the economic capability, personal taste and technology availability were very different.
- Secondly, we will collect a small but diverse set of data which will represent as many possible user types in purchase decision and how they interacted with the system in their purchase.
- We will then develop a special knowledge graph integrated with fuzzy cognitive maps to represent the causal relationship of the factors in the decision process. General knowledge graphs do not have decision making mechanisms to represent the trade-off of the conflicting factors.
- The initial structure of the knowledge graph will be constructed for the selected product with data and experts’ knowledge. Algorithms will be developed for the knowledge graph to be updated, through learning from data and user-system interaction.
- Algorithms will be developed for acquiring and analyzing the causal relationship knowledge underlying users’ purchase decision, which can be further applied to support their decision making. Similar users’ knowledge can be cross referred in the decision support process.
- Typical scenarios will be generated by applying the algorithms with either offline data or live data, to assess and illustrate the benefits of the new approach and algorithms.