Scientia potentia est. Knowledge is power.
Dataset
Our dataset contains +14k scientific articles from Scopus.
Recommenda works with scientific articles published between 2016 and the first half of 2018. This articles were published in the most relevant journals in the AI area. In fact, those were published on journals that were part of Q1 and Q2 in the JCR index. The articles were obtained through and ETL process developed with Python that used Scopus API as source. The dataset has a lot of useful information for researchers such as articles, authors, abstracts, keywords and even the different affiliations.
Topic extraction
NLP to retrieve the most relevant concepts.
In order to get more information about the similarities amongst different papers, we decided to extract different keywords from the articles. This keywords helps us find out the different clusters and research areas of our dataset. The topic extraction process may sound difficult at first, but the reality was different since there are implementations that make this task easier. In our case, Recommenda uses YAKE! algorithm to perform the topic extraction task.
Clustering
Identify different research areas amongst AI articles.
Clustering processes aim to find different groups in a dataset. In this case, we wanted to find out the different research areas on AI papers such as computer vision or fuzzy logic. After we performed the clustering process, we labeled the different groups based on the most relevant keywords from each cluster. Finally, we validated the results with an expert on this area. On the dataset tab you can explore the results of using hierarchical clustering algorithm to get the different groups. Users can filter the articles by year or by keywords.
Recommender System
Find the article that suits bests for your needs!
Recommenda trains and learns from the information of the articles of the dataset. It is able to make predictions and recommend similar articles based on its keywords. There are two ways to find your inspiration using Recommenda. On one hand, you can search for an specific article of our database and get the recommendations. On the other hand, you could introduce the title, abstract and keywords of a custom paper and get the 10 most similar articles to yours.