Multi-layer Search Relevance Scoring for Personal and Shared Content
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Efficient search for content in personal and shared databases is hindered by challenges such as remembering exact search terms, synonym, context, and homograph issues, leading to inconsistent results and reduced productivity, as traditional search methods fail to account for the unique characteristics of personal content collections like Evernote notebooks.
Innovation Solution
A system that determines partial relevance values for different characteristics of data subsets in personal and shared collections using term frequency and inverse document frequency, and displays relevant data subsets in order of relevance, incorporating multi-criteria similarity and filtering rules to identify related items within these collections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional search methods are used in personal and shared databases, then users can conduct basic text searches, but search accuracy and relevance deteriorate due to synonym, context, and homograph problems
Solution Approach 1:
The patent transforms the search process from simple keyword matching to a multi-parameter relevance scoring system. It calculates multiple relevance scores (r1, r2, r3, r4) based on different characteristics such as title matching, body text matching, tag matching, and source URL matching. This multi-dimensional parameter approach resolves the contradiction by considering multiple factors simultaneously to determine overall search relevance, thereby improving accuracy while preserving relevance information.
Solution Approach 2:
The patent segments the search relevance evaluation into distinct components: title relevance (r1), body text relevance (r2), tag relevance (r3), and source URL relevance (r4). Each component is calculated separately using appropriate algorithms (such as TF-IDF for text matching and exact matching for tags and URLs), then combined to form an overall relevance score. This segmentation allows the system to handle synonym, context, and homograph problems in different parts of the search query independently, improving overall search accuracy.
2Adaptability or versatility
If users search across diverse content collections including personal and shared databases, then the scope of information access increases, but search consistency and reliability deteriorate
Solution Approach 1:
The patent implements a universal search framework that works across diverse content collections including personal databases, shared databases, and web content. The same relevance scoring mechanism (calculating r1, r2, r3, r4 and combining them) is applied uniformly regardless of the source collection type. This multi-functional approach maintains search consistency and reliability while enabling users to search across multiple diverse collections simultaneously, resolving the contradiction between search scope and search consistency.
3Quantity of substance
If comprehensive search across multiple data collections is performed, then the quantity of searchable content increases, but search time and processing complexity increase
Solution Approach 1:
The patent implements a multi-layered search approach that performs partial searches at different levels. First, it searches personal databases using the full relevance scoring mechanism. Then, it performs a separate search in shared databases. Finally, it searches web content using a simplified approach. This partial action strategy allows the system to handle large volumes of searchable content across multiple collections while managing processing time by applying appropriate search depth to each collection type.
Solution Approach 2:
The patent segments the large-scale search task into separate searches across different data collections (personal database, shared database, web content). Each segmented search can be processed independently and in parallel, reducing the overall processing time compared to a monolithic search across all collections simultaneously. The segmentation also allows optimization of search parameters for each specific collection type.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Displaying relevant data subsets of at least one data collection includes providing an input data subset, determining a plurality of partial relevance values for different characteristics of each of the plurality of data subsets of the at least one data collection with respect to the input data subset according to a term frequency of terms in the input data subset and term in each of the plurality of data subsets of the at least one data collection and according to an inverse document frequency of the terms, determining a relevance of each of the plurality of data subsets of the at least one data collection based on the partial relevance values of the different characteristics, and displaying at least a portion of the data subsets of the at least one data collection in order of relevance.