# US 2016/0239572 A1 · Search engine classification

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

- Number: US 2016/0239572 A1
- Document ID: US20160239572A1
- Status: Published application
- Kind: A1
- Application number: 14/622,870
- Filed: 2015-02-15
- Published: 2016-08-18
- Earliest claimed: 2015-02-15
- Assignees: Microsoft Technology Licensing, LLC
- Applicants: Microsoft Technology Licensing, LLC
- Inventors: Alejandro Gutierrez Munoz; Jon Whisler; Adam Levi; Michael Golebiewski; Igor Rondel; Shahab Moradi
- Web page: https://patentgrep.com/patent/US20160239572A1

## CPC

- G06F 16/954: Navigation, e.g. using categorised browsing
- G06F 16/24578: {using ranking}
- G06F 16/285: {Clustering or classification}
- G06F 16/9535: Search customisation based on user profiles and personalisation
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Techniques for enabling a search engine to automatically classify the content type of Web documents. In an exemplary embodiment, Web documents may be classified as adult or non-adult, based on whether a document contains adult content. In an aspect, Web documents are mined offline to determine the presence of “adult hubs” to which adult documents are connected. The presence of such adult hubs is a strong indicator that linking Web documents may themselves contain adult content. Computational techniques for quantifying the connection between a candidate document and adult hubs are disclosed. The techniques may be utilized in an Internet search engine platform designed to accept user search queries and deliver highly relevant results.

## Claims

### Claim 1

A search engine apparatus for responding to a search query, the search engine apparatus comprising:
an adult hub detection block configured to calculate a hub score for an adult hub, wherein the adult hub is connected to a plurality of documents;
a classifier block comprising a candidate document feature block configured to calculate a document score for the candidate document, the classifier block configured to classify the candidate document as adult based on inputs comprising the document score of the candidate document and the hub score; and
a filter block configured to remove any candidate document classified as adult from responses to the search query.

### Claim 2 (depends on 1)

The apparatus of claim 1, the candidate document feature block comprising:
an n-gram analyzer configured to count a number of occurrences of target n-grams in any text portion of the candidate document; and
a media analyzer configured to detect the occurrence of explicit adult content in image or video present in the candidate document;
an aggregation block configured to aggregate the number of occurrences counted by the n-gram analyzer and the detected occurrences of explicit adult content.

### Claim 3 (depends on 1)

The apparatus of claim 1, further comprising:
a document feature block configured to calculate a document-level score for each of a plurality of documents in the network, said plurality of documents comprising at least one document connected to the candidate document in the network; and
a synthetic feature block configured to calculate at least one synthetic score for each of a plurality of document groups associated with the plurality of documents in the network, each synthetic score comprising an aggregation of a plurality of document scores.

### Claim 4 (depends on 3)

The apparatus of claim 3, at least one synthetic score comprising an aggregation of document-level scores over a single document group, at least one synthetic score comprising an aggregation of document scores over a single domain, and at least one synthetic score comprising an aggregation of document scores over a single host.

### Claim 5 (depends on 3)

The apparatus of claim 3, each document group comprising documents associated with a single container, the classifier block further comprising a hub distance score block configured to calculate a hub distance score measuring the hop distance between the candidate document and the adult hub.

### Claim 6 (depends on 5)

The apparatus of claim 5, the adult hub detection block further comprising:
a document aggregation block configured to aggregate document scores over all documents in the network connected to a hub candidate document;
a hub candidate classifier block configured to classify the hub candidate document as being an adult hub if said aggregated document scores exceeds a threshold.

### Claim 7 (depends on 5)

The apparatus of claim 5, the adult hub detection block further comprising:
a preliminary adult document classifier block configured to classify a document as a preliminary adult document if the corresponding document score exceeds a first threshold;
a preliminary adult group classifier block configured to classify a group as a preliminary adult group if the number of preliminary adult documents divided by a total number of documents in a document group exceeds a second threshold;
a hub candidate classifier block configured to classify a candidate hub as an adult hub if the number of preliminary adult groups divided by a total number of document groups associated with the candidate hub exceeds a third threshold.

### Claim 8 (depends on 7)

The apparatus of claim 7, wherein crowd-sourcing is utilized to verify the classification of at least one candidate hub as an adult hub.

### Claim 9 (depends on 5)

The apparatus of claim 5, the candidate document feature block further comprising:
a document classifier block configured to calculate a candidate document score corresponding to the candidate document;
a container classifier block configured to calculate a candidate container score corresponding to a container to which the candidate document belongs;
a domain classifier block configured to calculate a candidate domain score corresponding to a domain to which the candidate document belongs; and
a classifier aggregation block configured to aggregate the candidate document, candidate container, and candidate domain scores; the classifier block further comprising:
an adult status classification block configured to process the aggregated scores and the hub distance score using a classification algorithm to derive an adult status of the candidate document.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the classification algorithm is derived using machine learning techniques prior to receiving a user search request by the search engine.

### Claim 11

A method for classifying a candidate document using a search engine, the method comprising:
calculating a hub score for an adult hub, wherein the adult hub is connected to a plurality of adult documents;
calculating a document score for the candidate document; and
classifying the candidate document as adult based on inputs comprising the document score of the candidate document and the hub score.

### Claim 12 (depends on 11)

The method of claim 11, the calculating the document score comprising:
counting a number of occurrences of target n-grams in any text portion of the candidate document;
weighting each counted number by a target n-gram weighting factor; and
aggregating the weighted numbers to generate the document score.

### Claim 13 (depends on 11)

The method of claim 11, further comprising:
calculating at least one synthetic score for each of a plurality of document groups associated with a plurality of documents, each synthetic score comprising an aggregation of a plurality of document scores; and
if the candidate document is classified as adult, removing the candidate document from responses to a search query provided to the search engine.

### Claim 14 (depends on 13)

The method of claim 13, at least one synthetic score comprising an aggregation of document scores over a single document group, at least one synthetic score comprising an aggregation of document scores over a single domain, and at least one synthetic score comprising an aggregation of document scores over a single host.

### Claim 15 (depends on 13)

The method of claim 13, each document group comprising documents associated with a single container.

### Claim 16 (depends on 15)

The method of claim 15, further comprising:
aggregating document scores over all documents connected to a hub candidate document;
classifying the hub candidate document as being an adult hub if the aggregated document scores exceeds a threshold.

### Claim 17 (depends on 15)

The method of claim 15, further comprising:
classifying a document as a preliminary adult document if the corresponding document score exceeds a first threshold;
classifying a document group as a preliminary adult group if the number of preliminary adult documents divided by a total number of documents in a document group exceeds a second threshold;
classifying a candidate hub as an adult hub if the number of preliminary adult groups divided by a total number of document groups associated with the candidate hub exceeds a third threshold.

### Claim 18 (depends on 11)

The method of claim 11, the classifying the candidate document as adult comprising:
calculating a hub distance score measuring the hop distance between the candidate document and the adult hub;
calculating classifier scores corresponding to the candidate document, a container to which the candidate document belongs, and a domain to which the candidate document belongs;
aggregating the classifier scores; and
processing the aggregated classifier scores and the hub distance score using an algorithm to derive a status of the candidate document.

### Claim 19

A search engine apparatus for responding to a search query, the search engine apparatus comprising:
an adult hub detection block configured to calculate a hub score for an adult hub, wherein the adult hub is connected to a plurality of documents;
a classifier block comprising a candidate document feature block configured to receive a plurality of candidate documents, and calculate a document score for each candidate document, the classifier block configured to classify an adult status of each candidate document based on inputs comprising the document score of each candidate document and the hub score; and
a response block configured to return a candidate document classified as adult as a response to the search query, wherein the search query comprises a query for adult content.

### Claim 20 (depends on 19)

The apparatus of claim 19, the response block further configured:
to rank a plurality of candidate documents classified as adult using a composite score based on said inputs; and
to display the plurality of candidate documents in sequence according to rank based on the composite score.
