BACKGROUND
Success in any sort of online business is based on reputation. Receiving good or bad reviews about your products and services makes a difference.
Problem
Typical search engines search for links, images, videos but not for reviews about different products/services according to the buyer preferences. The buyer needs information about different aspects of the product s/he is interested in, not just millions of reviews ranked without any criteria. For instance, a buyer can specifically be interested in the battery life of a phone but not in its robustness.
Benefit
The correct selection of the reviews leads to time saved and wiser decisions when purchasing products or services in online businesses, thus leading to higher customer satisfaction.
METHODOLOGY & results
Architecture: On-premise development using local indices based on Lucene.
Developing language: Java
ML techniques: Unsupervised learning algorithms, fuzzy logic and NLP techniques.
Results: Depending on the complexity and the number of the opinions, the system can retrieve the most satisfying reviews with an accuracy between 80% and 90%.
