Recommended Datasets

Data offer recommmendations are generated via the PISTIS Marketplace Matchmaking Services. The Matchmaking Services are employed to link, as proactively as possible, data providers to data consumers, based on the latter’s interest in data assets (e.g., application domain), as well as complementary data assets, which can result in an increase of data value.

The service provides two main types of recommendations:

1. Content Similarity Recommendations

Everytime a Data Offer on the PISTIS Marketplace is viewed, a collection of similar data offers are presented as well. These can be found in the Similar Datasets section of the Offer Page.

These recommendations are informed purely by the characteristics of the offers and associated datasets. The content similarity recommender service analyzes descriptive metadata (such as title, keywords, categories, and domain), as well as structural and semantic properties of the datasets, to identify offers that are closely related.

By comparing these attributes across the marketplace, the service is able to construct pairwise similarity scores between every data offer. These scores are informed by all metadata characteristics of both the dataset and the published offer.

This approach ensures that recommendations remain relevant regardless of user history or activity.

2. User Interactions-Based Recommendations

Notice that the prior set of recommendations are agnostic of the user viewing the data offer. In contrast, user interactions–based recommendations are personalized and tailored to each individual user. These can be found in the Data Datasets section of the Offer Page.

These recommendations are generated by analyzing how users interact with data offers across the PISTIS Marketplace. From these interactions, the matchmaking service infers user preferences and uses this information to recommend relevant datasets.

Unlike content similarity recommendations, these suggestions evolve dynamically as more interactions are recorded, ensuring that the recommendations remain up-to-date and increasingly accurate over time.

Together, content-based and interaction-based recommendations provide a comprehensive discovery experience, supporting both context-aware exploration and personalized dataset discovery within the PISTIS Marketplace. These recommendations are kept fresh via regular retraining schedule.