Semantic Vector Encoding and Similarity Search Using Fulltext Search Engines
Proceedings of the 2nd Workshop on Representation Learning for NLP
Vector representations and vector space modeling (VSM) play a central role in modern machine learning. We propose a novel approach to ‘vector similarity searching’ over dense semantic representations of words and documents that can be deployed on top of traditional inverted-index-based fulltext engines, taking advantage of their robustness, stability, scalability and ubiquity. We show that this approach allows the indexing and querying of dense vectors in text domains. This opens up exciting avenues for major efficiency gains, along with simpler deployment, scaling and monitoring. The end result is a fast and scalable vector database with a tunable trade-off between vector search performance and quality, backed by a standard fulltext engine such as Elasticsearch. We empirically demonstrate its querying performance and quality by applying this solution to the task of semantic searching over a dense vector representation of the entire English Wikipedia.
Similarity Ranking as Attribute for Machine Learning Approach to Authorship Identification
Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)
In the authorship identification task, examples of short writings of N authors and an anonymous document written by one of these N authors are given. The task is to determine the authorship of the anonymous text. Practically all approaches solved this problem with machine learning methods. The input attributes for the machine learning process are usually formed by stylistic or grammatical properties of individual documents or a defined similarity between a document and an author. In this paper, we present the results of an experiment to extend the machine learning attributes by ranking the similarity between a document and an author: we transform the similarity between an unknown document and one of the N authors to the order in which the author is the most similar to the document in the set of N authors. The comparison of similarity probability and similarity ranking was made using the Support Vector Machines algorithm. The results show that machine learning methods perform slightly better with attributes based on the ranking of similarity than with previously used similarity between an author and a document.