# US 10,482,146 B2 · Systems and methods for automatic customization of content filtering

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

- Number: US 10,482,146 B2
- Document ID: US10482146B2
- Status: Granted
- Kind: B2
- Application number: 15/591,797
- Filed: 2017-05-10
- Published: 2018-02-15
- Granted: 2019-11-19
- Earliest claimed: 2016-05-10
- Assignees: Massachusetts Institute of Technology
- Applicants: Massachusetts Institute of Technology
- Inventors: Jason R. Thornton; Marianne A. Deangelus; Benjamin Andrew Miller; Kevin Kyung Nam
- Web page: https://patentgrep.com/patent/US10482146B2

## CPC

- G06F 16/9535: Search customisation based on user profiles and personalisation
- G06N 20/00: Machine learning
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Exemplary embodiments include a system and method for Automatic Customization of Content Filtering (ACCF). The method includes receiving a search string from a user, creating a first filter based on the search string and searching different types of content stored in different content locations based on the first filter. The search returns a first subset of results to the user based on the first filter. The method further receives an indication of relevance for each one of the results in the first subset from the user. The method dynamically creates a second filter based on the received indications of relevance for the first subset of results. Based on the second filter, a second subset of more relevant results is returned to the user.

## Claims

### Claim 1

A computing-device implemented method for customization of content filtering, comprising:
receiving a search string from a user;
creating a first filter based on the search string;
performing a first search of a plurality of modalities of content stored in a plurality of locations based on the first filter, the plurality of modalities of content including at least two of text, image and video content;
returning a first subset of results from the first search to the user;
receiving from the user an indication of relevance or non-relevance for each one of the results in the first subset;
dynamically creating a second filter by:
creating a plurality of feature spaces by extracting a plurality of features from the plurality of modalities of content referenced by the first subset of results, the extracted features including at least two types of text features including two or more of N-Grams, latent topics, parts-of-speech, contextual content flags and contextual metadata, and at least two types of image or video features including two or more of intensity and color histograms, bag-of-words scale-invariant feature transform (SIFT), border/interior classification, dense SIFT, GIST, histogram orientation gradients HOG, a number of faces, and a pedestrian confidence score;
reducing a dimensionality of each feature space in the plurality of feature spaces; and
applying a classifier trained on the user indication of relevance or non-relevance to the reduced feature spaces; and
returning a second subset of results to the user generated by applying the classifier.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
receiving an indication of relevance for each one of the results in the second subset of results from the user;
dynamically adjusting the second filter based on the received indications of relevance for the results in the second subset; and
returning a third subset of results from a third search based on the first filter and the adjusted second filter.

### Claim 3 (depends on 2)

The method of claim 2, further comprising:
iteratively:
receiving an indication of relevance from the user for each result in an additional subset of results generated by an additional search of a plurality of modalities of content;
further adjusting dynamically the second filter based on the received indications of relevance for the results in the additional subset; and
returning another subset of results based on the first and the further adjusted second filter,
wherein iterations continue until receiving a user indication of acceptable results.

### Claim 4 (depends on 1)

The method of claim 1, wherein extracting features from a text modality includes extracting term frequencies.

### Claim 5 (depends on 1)

The method of claim 1, further comprising reducing a feature space by removing redundant features.

### Claim 6 (depends on 1)

The method of claim 1, wherein the feature space is reduced using one or more of unsupervised techniques and supervised techniques.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
comparing, with the classifier, extracted features of a modality to features from a plurality of modalities from the first subset of modalities based on the user feedback on the first set of modalities.

### Claim 8 (depends on 7)

The method of claim 7, further comprising:
issuing, via the classifier, a score for each extracted feature based on the comparison of the extracted features of a modality to features from a plurality of modalities from the first subset of modalities.

### Claim 9

A computing system for customization of content filtering, comprising:
a computing system including a server operatively coupled to a plurality of content storage locations holding a plurality of modalities of content including at least two of text, image and video content, the computing system programmed to:
receive a search string from a user;
create a first filter based on the search string;
perform a first search of a plurality of modalities of content stored in a plurality of locations based on the first filter;
return a first subset of results from the first search to the user;
receive from the user an indication of relevance for each one of the results in the first subset;
dynamically create a second filter by:
creating a plurality of feature spaces by extracting a plurality of feature from the plurality of modalities of content referenced by the first subset of results, the extracted features including at least two types of text features including two or more of N-Grams, latent topics, parts-of-speech, contextual content flags and contextual metadata, and at least two types of image or video features including two or more of intensity and color histograms, bag-of-words scale-invariant feature transform (SIFT), border/interior classification, dense SIFT, GIST, histogram orientation gradients HOG, a number of faces, and a pedestrian confidence score;
reducing a dimensionality of each feature space in the plurality of feature spaces; and
applying a classifier trained on the user indication of relevance or non-relevance to the reduced feature space; and
return a second subset of results to the user generated by applying the classifier.

### Claim 10 (depends on 9)

The system of claim 9, wherein the computing system is programmed to:
receive an indication of relevance for each one of the results in the second subset of results from the user;
dynamically adjust the second filter based on the received indications of relevance for the results in the second subset; and
return a third subset of results from a third search based on the first filter and the adjusted second filter.

### Claim 11 (depends on 10)

The system of claim 10, wherein the computing system is programmed to:
iteratively:
receive an indication of relevance from the user for each result in an additional subset of results generated by an additional search of a plurality of modalities of content;
further adjust dynamically the second filter based on the received indications of relevance for the results in the additional subset; and
return another subset of results based on the first and the further adjusted second filter,
wherein iterations continue until receiving a user indication of acceptable results.

### Claim 12 (depends on 9)

The system of claim 9, wherein a feature space is reduced by removing redundant features.

### Claim 13 (depends on 12)

The system of claim 12, wherein the feature space is reduced using one or more of unsupervised techniques and supervised techniques.