Interactive Similarity Search
The meaningful definition of similarity is key for many Machine Learning, Data Mining, or Information Retrieval approaches. GRIS goes a visual way to search for meaningful similarity functions. With the use of visual-interactive interfaces, users are enabled to define their notion of similarity, which is then learned by a similarity learning model. Our vision is that in future, users will be able to define their similarity measure before they use information retrieval, data mining and machine learning technology.
Information Visualization and Visual Analytics techniques help to design visual interfaces that support users in the expression of their notion of similarity. In recent works enabled user to, e.g., define the similarity of countries, or the similarity between pairs of soccer players. The result is a model that represents the learned notion of the users’ similarity. Learning similarity functions builds upon technique from data mining and machine learning. As such, Interactive Similarity Search is an example for the interactive machine learning process.
Interactive Similarity Search explained
Interactive Similarity Search addresses the problem of searching for meaningful similarity measures (similarity search) in a visual and interactive way. Users play an active role in the definition of similarity. The result is a similarity measure (~ a distance function) that can be used in many information retrieval, data mining, and machine learning environments. In particular, unsupervised data analysis tasks can be supported with similarity measures.
Interactive Similarity Search also allows the user-centered definition of similarity measures: in a training phase (where the measure is learned) users can express their personal notion of similarity. This notion is learned by a similarity learning model and can be used for downstream data analysis tasks.