Neural Network Entity Prediction for Multi-Tenant Search

Solution Overview

Problem

In large database systems, particularly multi-tenant CRM systems, predicting the relevant entity type for search query results is challenging due to the vast number of records and 'false' hits, leading to a negative user experience and high resource consumption.

Innovation Solution

A neural network model is implemented to predict the most relevant entities for search query results by training on user search patterns, character, word, frequency view, and organization encodings, enhancing the user experience by providing accurate and efficient search results tailored to each organization's specificities.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional search methods are used in large database systems, then comprehensive search results can be returned, but system resources are consumed heavily and response time is long

Engineering Contradiction:

Improvesearch result completenessVSAvoidsearch processing speed

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The system performs preliminary actions by pre-computing and storing entity embeddings and relationship data in advance. When a search query is received, the system leverages these pre-computed structures to rapidly retrieve and rank relevant entities, avoiding the need to process the entire database from scratch and thus achieving fast response times without sacrificing result completeness

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary layer consisting of entity embeddings and graph-based relationship structures that mediate between the raw database and the search query. This intermediary representation allows the system to efficiently filter and rank entities based on their semantic similarity and relationship strength, reducing the computational burden while maintaining comprehensive search capabilities

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If traditional search methods are used in large database systems, then all matching records can be returned, but many 'false' hits are included and relevant information is buried

Engineering Contradiction:

Improvesearch result completenessVSAvoidsearch result relevance

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The patent introduces an intermediary layer consisting of entity embeddings and graph-based relationship structures that mediate between the raw database and the search query. This intermediary representation allows the system to efficiently filter and rank entities based on their semantic similarity and relationship strength, reducing the computational burden while maintaining comprehensive search capabilities

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes the parameter space by transforming entity attributes into continuous embedding vectors and using similarity metrics (e.g., cosine similarity) instead of traditional exact matching. This transformation enables the system to rank entities by relevance score, pushing false hits to the bottom of the result list and surfacing the most relevant information at the top

Inventive Principle:

Principle #35Parameter changes

3Reliability

If comprehensive search of all records is performed, then complete results are obtained, but user experience becomes negative due to long processing time

Engineering Contradiction:

Improvesearch result completenessVSAvoidsearch response time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by pre-computing and storing entity embeddings and relationship data in advance. When a search query is received, the system leverages these pre-computed structures to rapidly retrieve and rank relevant entities, avoiding the need to process the entire database from scratch and thus achieving fast response times without sacrificing result completeness

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent segments the search process into distinct stages: query embedding, graph traversal based on relationship types, and ranked result generation. This segmentation allows the system to process only the relevant portion of the database corresponding to the query intent, significantly reducing processing time while maintaining comprehensive results

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS10970336B2Predicting entities for search query results

Publication Date: 2021.04.06 SALESFORCE INC

AI summary

For a database accessible by a plurality of separate organizations, a system is provided for predicting entities for database query results. The system includes a multi-layer neural network. The system is configured to receive a query encoding for one or more previous queries made into the database, a user entity view frequency encoding for a frequency of views by one or more users, and an organization encoding for one or more separate organizations accessing the database; and based on the query encoding, the user entity view frequency encoding, and the organization encoding, generate a neural model for predicting entities for results to a present query into the database. In some embodiments, the neural model is global across the separate organizations accessing the database.