Improving Search and Recommendation for e-Commerce
Investigating ‘cold-start’ product-database search and recommendation.
In e-commerce, it is of central importance to present the products users are looking for as soon as possible. Products that are not presented, even though a consumer would be interested in them, result in unexploited revenue for the company. More importantly, products that are presented – even though the consumer is not interested in them – lead to eroded trust among customers.
Of particular difficulty are the so-called dynamicity and cold-start problems. Cold-start denotes the situation where there is a lack of information with respect to the user or the product’s popularity. State-of-the-art research into recommendation shows that accounting for dynamicity with respect to drifting popularity and user interests leads to improvements in system performance.
This research project outlines real-world investigation of the vital areas of coldstart product database search and recommendation. It turns out that the problem that e-commerce sites face in the cold-start can be modelled by a well-known two-player game called the Contextual Bandit problem, which asks for the best strategy of product retrieval that the site can make given the user’s query