# US 2023/0409614 A1 · Search analysis and retrieval via machine learning embeddings

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0409614 A1
- Document ID: US20230409614A1
- Status: Published application
- Kind: A1
- Application number: 17/971,491
- Filed: 2022-10-21
- Published: 2023-12-21
- Earliest claimed: 2022-06-15
- Assignees: UnitedHealth Group Incorporated
- Applicants: UnitedHealth Group Incorporated
- Inventors: Laura D. Hamilton; Vinit Garg; Ayush Tomar; Martin R. Linenweber; Preet Kamal S. Bawa; David Armbrust; Rupesh Kartha; Lun Yu
- Web page: https://patentgrep.com/patent/US20230409614A1

## CPC

- G06F 16/3325: {Reformulation based on results of preceding query}
- G06F 16/9024: {Graphs; Linked lists (G06F16/9027 takes precedence)}
- G06F 16/3334: {Selection or weighting of terms from queries, including natural language queries}
- G06F 16/387: using geographical or spatial information, e.g. location
- G06F 40/30: Semantic analysis
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/08: Learning methods
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Various embodiments of the present disclosure provide methods, apparatus, systems, computing devices, computing entities, and/or the like for retrieving relevant items for user queries by generating, using a search engine machine learning model, a prediction-based action for the query input wherein query input embeddings of the query input are generated. For each query input embedding, a k-Nearest-Neighbor (KNN) search is performed with respect to search engine repository item embeddings to generate initial search results, and for each initial set result, performing N hops within a semantic graph starting from nodes associated with the initial search result to generate related search results. The search engine machine learning model is trained by generating a search engine repository item embeddings according to embedding techniques for respective content categories and generating the semantic graph based at least in part on a measure of similarity for pairs of search engine repository item embeddings.

## Claims

### Claim 1

A computer-implemented method, in a data processing system comprising a processor and a memory, for retrieving relevant items for user queries, the computer-implemented method comprising:
receiving, by a computing device, a query input;
generating, by the computing device and using a search engine machine learning model, a prediction-based action for the query input wherein:
for each content category, a query input embedding of the query input is generated with respect to the content category;
for each query input embedding, a k-Nearest-Neighbor (KNN) search is performed with respect to search engine repository item embeddings to generate initial search results;
for each initial set result, performing N hops within a semantic graph starting from nodes associated with the initial search result to generate related search results;
training, by the computing device, the search engine machine learning model, the training comprising:
for each search engine repository item of a search engine repository, assigning one or more content category labels and generating a search engine repository item embedding according to embedding techniques for respective content categories associated with the one or more content category labels; and
generating the semantic graph based at least in part on a measure of similarity for pairs of search engine repository item embeddings associated with pairs of the search engine repository items.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein a search engine repository item includes metadata corresponding to the one or more content category labels.

### Claim 3 (depends on 1)

The computer-implemented method of claim 1, wherein the respective content categories comprise at least one of: syntactic content, semantic content, and geospatial content.

### Claim 4 (depends on 1)

The computer-implemented method of claim 1, wherein the embedding techniques comprise syntactic embedding by:
tokenizing content of a given search engine repository item with subword n-grams;
training a term-frequency inverse document frequency (TF-IDF) model on the tokenized content; and
removing bottom tokens with low TF-IDF values from the tokenized content.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, wherein the embedding techniques comprise semantic embedding by:
retrieving a domain corpus; and
tuning a universal sentence encoding model on the domain corpus.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, wherein the embedding techniques comprise geohash embedding by:
converting geospatial information into one or more geocodes according to a geohashing algorithm;
identifying, from a data source, procedure geocode pairs based at least in part on procedures associated with respective ones of the one or more geocodes; and
training a skip-gram model based at least in part on the identified procedure geocode pairs.

### Claim 7 (depends on 1)

The computer-implemented method of claim 1, wherein generating the semantic graph further comprises:
retrieving semantic embedding for each of the search engine repository items;
computing the measure of similarity;
generating edges between the pairs of search engine repository items in the semantic graph based at least in part on the computed measure of similarity; and
removing given ones of the edged according to a threshold value.

### Claim 8 (depends on 1)

The computer-implemented method of claim 1, further comprising generating a user profile and history embedding based at least in part on historical information and activities data.

### Claim 9 (depends on 1)

The computer-implemented method of claim 1, wherein generating the one or more search results further comprises computing a user relevance score for ranking the one or more search results.

### Claim 10 (depends on 9)

The computer-implemented method of claim 9, wherein the user relevance score is computed based at least in part on semantic relevance, syntactic relevance, geospatial relevance, and user relevance.

### Claim 11 (depends on 1)

The computer-implemented method of claim 1, wherein the measure of similarity comprises pairwise cosine similarity.

### Claim 12

An apparatus for retrieving relevant items for user queries, the apparatus comprising at least one processor and at least one memory including program code, the at least one memory and the program code configured to, with the processor, cause the apparatus to at least:
receive a query input;
generate, using a search engine machine learning model, a prediction-based action for the query input wherein:
for each content category, a query input embedding of the query input is generated with respect to the content category;
for each query input embedding, a k-Nearest-Neighbor (KNN) search is performed with respect to search engine repository item embeddings to generate initial search results;
for each initial set result, performing N hops within a semantic graph starting from nodes associated with the initial search result to generate related search results; and
train the search engine machine learning model, the training comprising:
for each search engine repository item of a search engine repository, assigning one or more content category labels and generating a search engine repository item embedding according to embedding techniques for respective content categories associated with the one or more content category labels; and
generating the semantic graph based at least in part on a measure of similarity for pairs of search engine repository item embeddings associated with pairs of the search engine repository items.

### Claim 13 (depends on 12)

The apparatus of claim 12, wherein a search engine repository item includes metadata corresponding to the one or more content category labels.

### Claim 14 (depends on 12)

The apparatus of claim 12, wherein the respective content categories comprise at least one of: syntactic content, semantic content, and geospatial content.

### Claim 15 (depends on 12)

The apparatus of claim 12, wherein the embedding techniques comprise syntactic embedding by:
tokenizing content of a given search engine repository item with subword n-grams;
training a term-frequency inverse document frequency (TF-IDF) model on the tokenized content; and
removing bottom tokens with low TF-IDF values from the tokenized content.

### Claim 16 (depends on 12)

The apparatus of claim 12, wherein the embedding techniques comprise semantic embedding by:
retrieving a domain corpus; and
tuning a universal sentence encoding model on the domain corpus.

### Claim 17 (depends on 12)

The apparatus of claim 12, wherein the embedding techniques comprise geohash embedding by:
converting geospatial information into one or more geocodes according to a geohashing algorithm;
identifying, from a data source, procedure geocode pairs based at least in part on procedures associated with respective ones of the one or more geocodes; and
training a skip-gram model based at least in part on the identified procedure geocode pairs.

### Claim 18 (depends on 12)

The apparatus of claim 12, wherein generating the semantic graph further comprises:
retrieving semantic embedding for each of the search engine repository items;
computing the measure of similarity;
generating edges between the pairs of search engine repository items in the semantic graph based at least in part on the computed measure of similarity; and
removing given ones of the edged according to a threshold value.

### Claim 19 (depends on 12)

The apparatus of claim 12, further comprising causing the apparatus to:
generate a user profile and history embedding based at least in part on historical information and activities data.

### Claim 20

A computer program product for retrieving relevant items for user queries, the computer program product comprising at least one non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions configured to:
receive a query input;
generate, using a search engine machine learning model, a prediction-based action for the query input wherein:
for each content category, a query input embedding of the query input is generated with respect to the content category;
for each query input embedding, a k-Nearest-Neighbor (KNN) search is performed with respect to search engine repository item embeddings to generate initial search results;
for each initial set result, performing N hops within a semantic graph starting from nodes associated with the initial search result to generate related search results; and
train the search engine machine learning model, the training comprising:
for each search engine repository item of a search engine repository, assigning one or more content category labels and generating a search engine repository item embedding according to embedding techniques for respective content categories associated with the one or more content category labels; and
generating the semantic graph based at least in part on a measure of similarity for pairs of search engine repository item embeddings associated with pairs of the search engine repository items.
