Voice Queries in Product Search with Vague Conditions
2019-04-01

Teaching personnel


Lecturer

Formalia


Targeted audience
  • Angewandte Informatik Bachelor
  • Angewandte Informatik Master
  • Komedia Bachelor
  • Komedia Master
Preconditions
  • Computer Scientists: Advanced programming skills & Machine Learning (desired)
  • Komedia: UI Design & Evaluation, Knowledge about IR (desired)

Description


Traditional Product Search on E-Commerce Platforms like Amazon is provided by the Web domain. Users can search for products either by text queries or facet search. However, the emerging increase of personal assistants like Amazon Alexa allows us faster shopping activities without typing. The advantage of speaking faster than typing leads to the prediction that voice search will replace web search in any time soon. By 2020 50% of all searches will be voice searches. Voice Product Search differs from Web Product Search in terms user behaviour. Voice assistants provide users buying or rejecting options on single products. In the classic web shopping domain the user can navigate through a set of presented products. Unlike Voice Shopping it is more likely that a user perform more navigation actions before buying a product in the case of Web Search.

In this project the aim is to bridge Web Search and Voice Search in the use case of Product Search. Users should have the oppurtunity to perform natural spoken critiques on specific products according to selected attributes so that the system outputs products that fit to the user's preferences. In this fashion, the user can navigate through the assortment and hopefully find the "best match" product. Based on a User Interface it is possible to point on interesting products and to perform voice queries. Another problem arises with vague conditions. Users often have vague and limited knowledge of what they are searching for. For example, someone is searching for a "high performance laptop" allows many candidate items to investigate on. It should be possible to handle vague query conditions and thus not excluding possible relevant products.

The project aims to have up to 15 students with strong programming (e.g. Java or Python) and machine learning backgrounds (e.g. IM exam). It is also important that the student is familiar with the concept of IR, search engines and UI Design & Evaluation.