Optimizations of search engines for merging search results

Inventors

Lachmi, Karin • Levitt, Daniel • Grunspan, Dan • Pardo, Ehud

Assignees

Bioz Inc

Abstract

Disclosed herein are methods, systems, and apparatuses, including computer programs encoded on computer storage media, for computing numeric representations of words. One of the methods includes obtaining, from a corpus of journals, a plurality of objects. The corpus is analyzed for contextual clues as to the nature of each of these objects. Contextual analysis may include an analysis of particular fields of metadata to do with the objects. Another example of contextual analysis involves identification of an experiment that a group of objects are commonly associated with and merging objects therefrom. A third contextual analysis example makes use of word vectors about each object. Using these contextual analyses, objects are merged together with the context of search results of a search engine.

Core Innovation

A natural language processor extracts a plurality of objects from a corpus of text, where each object is a term within a document that contextually identifies a usable item. The natural language processor attaches context data to objects, including a source associated with each respective object and found within the corpus of text. The objects are integrated as separate searchable results within a search engine before any merging is performed.

The search engine merges the first object and the second object into a single searchable result based on the first object and the second object having corresponding context data. In one approach, merging is based on generating a first word vector and a second word vector for the objects from words neighboring each respective object in the corpus of text and merging when a similarity threshold is exceeded. In another approach, merging is based on certainty values or on matching associated experiment information where both objects share the same associated experiment.

The system architecture supports this merging as a search-engine optimization that uses semantic processing submodules including tokenization, chunking, relations, binding, and extraction. It includes document storage and a data pipeline with back-end processing and front-end search-engine components, and it integrates the resulting merged searchable results for downstream user presentation.

Claims Coverage

The independent claims cover methods and a system that extract context-identifying term objects from a corpus of text using a natural language processor, integrate them as separate searchable results in a search engine, and merge corresponding objects into a single searchable result using context-derived correspondence. Across the independent claims, the inventive features focus on context-data attachment, merge criteria based on corresponding context data, and merge criteria based on experiment matching or word-vector similarity (with a predefined similarity threshold).

Overall, the independent claims require extracting term objects that contextually identify usable items, attaching context data including a source, integrating objects as separate searchable results, and merging them into a single searchable result when corresponding context data indicates correspondence. They also cover alternative merge criteria including word-vector similarity exceeding a predefined threshold and, in dependent claim coverage, merge criteria based on shared associated experiment and/or certainty thresholds.

Stated Advantages

Documented Applications