An efficient keyword recommendation Service for a leading society publisher

IFIS (Food-focused Information Search Tools & Research) is a not-for-profit publisher & educational charity helping researchers & students find scientific food information they can trust & build their research skills.

Challenges:

Keyword based search dominates the internet. IFIS was looking to enhance their services for its users and aimed at providing appropriate keyword recommendations to power their research and boost visibility. Often academic researchers wonder which thesaurus terms or keywords to use and what are the right keywords for their research or publication needs. In order to address this need, IFIS began search for a partner with a high repute in deep learning and AI technologies and who could analyze the entire Food Science and Technology Abstracts (FSTA) database of scholarly articles in the fields of Food and Health Science and suggest keywords that match their title and abstract.

Solution:

Utilizing semantic similarity algorithm, the Molecular Connections team developed a custom-built, high throughput, retainable and multifaceted algorithm by exploiting the recent advances in deep learning to provide a flexible solution that can cater to similar contexts.

The system returns keyword data from the top 100 similar articles identified by an artificial intelligence algorithm. It displays the frequency of occurrence of keywords used within these top 100 similar articles.

The team incorporated various contextual vectorizations/representations: word and sentence embedding in conjugation; custom inserted generalized layers across the model; bi-directional transformers were used as an attempt to construct and learn dependency grammars; and a Siamese encoder-decoder was used in conjugation with a BERT: a deep bi-directional transformer to handle a large corpus of information and ensure that it works efficiently with live inputs.

To ensure optimum performance, the team also employed few techniques such as Quantization and Approximate Nearest Neighbours (ANN) search. KRS Semantic similarity is built on state-of-the-art deep learning algorithms with due diligence to intended application and performance metrics.

Benefits:

The client was able to achieve optimum result from the project. The KRS Semantic resulted in better submission, and it not only provided relevant and reliable search results leading to impactful content and promotion, but also helped users with SEO. There was a 40 percent jump in traffic.