"A machine can be said to be intelligent if its answers are indistinguishable from those of a human being" Alan Turing
- Hybrid Content-Based & Collaborative Recommendation System
- Action (Clicks, Likes, Dislikes, etc.) Prediction System
- Context-Dependent Recommendation System
- Cross-Domain Recommendation System
- User Engagement & Content Diversity Optimization System
- Context-Aware Notification System
- Next Place Prediction System
- Relevant Email and Contacts Classification System
- Advanced Statistical and Deep Learning Techniques
- Text Summarization
- Duplicated and Related News Identification
- Action Required Email Classification
- Fake News Detection
- Federated Learning
- Life-Long Learning
- Among others...
The AI engine constantly works to analyze all the information and to always have the most relevant information available, without the need to perform searches. Our algorithms are capable of detecting underlying patterns in the data and give coherence to the information coming from different domains, a priori unrelated. The following are some examples of the capabilities of the AI engine:
- Relevant personalized recommendations from the start: The AI engine is able to characterize the preferences of a new user by integrating the knowledge of the preference patterns of the existing user population, with minimal information provided by the user, through the use of Bayesian Networks. Our models are able to offer stable user profiles in cold start situations.
- The advanced Natural Language Processing models allow us to characterize and evolve the preferences of the users with various levels of abstraction, dynamically integrating highly detailed linguistic information with general interest topics.
- Our Opinion Mining & Sentiment Analysis algorithms permit the extraction of subjective information about the content and the ability to offer personalized recommendations based on the location of the user and their habitual response to the emotional charge of the content.
- The processing of geo-location signals and the interaction of the user with the platform through advanced Unsupervised Classification and Structured Semantic Analysis techniques allows us to understand the user’s various contexts and adapt the recommendations to said contexts. Plus, a combination of Bayesian models and computational Machine Learning models allows us to predict changes in the user’s context, such as if the user is going to go to a different location soon.
- The AI engine has multiple recommendation systems adapted to the various needs of the user and the information domains. Similarly, it has general purpose recommendation systems based on hybrid models (content and collaborative-based) and on Machine Learning (Action Prediction Model).
Sherpa is leading the way in the research and development of machine learning techniques for intelligent predictive assistants, and we are paving the way for novel applications that respect users’ privacy, based on cutting edge research on Federated Learning.
Francisco Herrera, Ph.D.
Senior Associate Researcher in DL & ML of Sherpa.ai.
Highly Cited Researchers (Thomson Reuters) in the areas of Engineering and Computer Sciences.
Spanish National Award on Computer Science.
The algorithms developed by Sherpa are capable of predicting the future of our users, before they're even aware of it themselves.
JOSE A. LOZANO, Ph.D.
Algorithms & Models Senior Associate Researcher of Sherpa.ai
Ph.D. in Computer Science.
Degree in Mathematics & M.Sc.
Associate editor of top journals, such as IEEE Transactions on Neural Networks and Learning Systems and IEEE Trans. on Evolutionary Computation.
Several best paper awards from international conferences such as the World Conference on Computational Intelligence and the IEEE Congress on Evolutionary Computation.
Our models incorporate and apply the most recent advances in Machine Learning to Natural Language Processing and the Dialogue Manager.
Deep Learning Models (Recurrent Neural Networks, Attentional Mechanism, Encoder-Decoders, etc.) and Reinforcement Learning allow us to extract the relevant information from the text documents and offer our users content that is more relevant for them:
- Linguistic Analysis
- Word Embeddings
- Opinion Mining & Sentiment Analysis
- Stance Detection
- Fake News Detection
- Identification of Duplicate and Related Documents
- Automatic Summaries
Sherpa incorporates five levels of linguistic analysis to eliminate all possibility of misinterpretation - morphological, syntactical, semantic, pragmatic, and functional. Its sophisticated natural language technology mimics human understanding to dismiss impossible or unlikely matches.
Packed with over 300,000 concepts and 5,000 syntactic and semantic rules, Sherpa's thoroughly tested core system provides the basis for a reliable and comprehensive approach to human-computer interaction.
At Sherpa.ai research lab we are working on the next generation of assistants using the latest Machine Learning paradigms, such as Reinforcement Learning and Life-Long Learning.
ENEKO AGIRRE, Ph.D.
Senior Associate Researcher in NLP of Sherpa.ai
Ph.D. Computer Science.
Google Research Awards in 2016 and 2018.
Sherpa’s extensive resources of conceptual and linguistic information and its detailed, five-level approach to linguistic analysis makes it a highly accurate and flexible tool for building innovative natural language-based solutions.
DEBORAH DAHL, Ph.D.
Speech and Natural Language Processing Expert
Co-Principal Investigator on the Defense Advanced Research Projects Agency (DARPA) of the U.S. Department of Defense-funded project which integrated Unisys natural language understanding technology with speech recognition.