# US 2021/0241163 A1 · Establishing a communication session between client terminals of users of a social network selected using a machine learning model

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

- Number: US 2021/0241163 A1
- Document ID: US20210241163A1
- Status: Published application
- Kind: A1
- Application number: 16/777,915
- Filed: 2020-01-31
- Published: 2021-08-05
- Earliest claimed: 2020-01-31
- Assignees: salesforce.com, inc.
- Applicants: salesforce.com, inc.
- Inventors: Noa WITZTUM; Ofer Bachner; Dima Volski
- Web page: https://patentgrep.com/patent/US20210241163A1

## CPC

- G06N 20/00: Machine learning
- G06F 16/285: {Clustering or classification}
- G06Q 10/10: Office automation; Time management
- G06Q 10/42: Determination of affinities or common interests between users
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

There is provided a method, comprising: extracting user feature profiles for users of a social network, each feature profile being structured and including user features extracted from unstructured user generated text, indications of participation in groups, and structured user profiles, training a clustering-component of a model to cluster the feature profiles, training a matching-component of the model to compute a distance score indicative of statistical similarity between a feature profile of a target user and features profiles of other users of a same cluster, using a training dataset of pairs of feature profiles extracted from common clusters, each pair assigned a distance score label, providing the model for: identifying a certain cluster of a certain user, and computing distance scores between the feature profile of the certain user and other feature profiles of other users of the certain cluster for selecting one user for establishment of a communication session.

## Claims

### Claim 1

A computer implemented method for training a machine learning model for establishing a communication session, comprising:
extracting a plurality of user feature profiles for a plurality of users of a social network hosted by at least one social network server, each user feature profile being structured and including a plurality of user features extracted from unstructured user generated text, indications of participation in groups, and structured user profiles;
training a clustering-component of the model to cluster the plurality of user feature profiles;
training a matching-component of the model to compute a distance score indicative of statistical similarity between a feature profile of a target user and features profiles of other users of a same cluster, using a training dataset of pairs of user feature profiles extracted from common clusters, each pair assigned a distance score label;
providing the model for:
identifying a certain cluster of a certain user; and
computing distance scores between the user feature profile of the certain user and other user feature profiles of other users of the certain cluster for selecting a subset of the other users;
wherein the communication session is established between an initiating client terminal of the certain user and the target client terminal of a user selected by the initiating client terminal from the subset.

### Claim 2 (depends on 1)

The method of claim 1, wherein the distance score label of each pair is computed as a statistical distance indicative of correlation score between the feature profiles of the users of the respective pair, or assigned a value denoting a maximal distance score provided by one of the users of the respective pair, or assigned a value denoting minimal distance score provided by one of the users of the respective pair.

### Claim 3 (depends on 2)

The method of claim 2, further comprising, after establishing the communication session, obtaining an indication of the distance score from the initiating client terminal for update of the label of the respective pair, and updating the matching-component of the model.

### Claim 4 (depends on 3)

The method of claim 3, further comprising:
receiving from the initiating client terminal, an indication of no selection of the subset of the other users included in the certain cluster;
updating labels of pairs of the certain user and each of the subset of the other users with an indication of a maximal distance score;
updating the clustering-component and matching-component of the model with the updated labels of the pairs;
computing updated clusters by the updated clustering-component;
computing updated distance scores by the updated matching-component; and
receiving a new selection of a new subset of other users included in a new certain cluster presented on the initiating client terminal.

### Claim 5 (depends on 1)

The method of claim 1, wherein the clustering-component is trained to cluster user feature profiles according to a selected target subset of user features and to not cluster the user feature profiles according to a selected variable subset of user features.

### Claim 6 (depends on 5)

The method of claim 5, wherein the target subset of the user features are selected from the group consisting of: words extracted from an employee title, department, business unit, and country, user identification profile, ranking in the social network, membership in groups of the social network, and words extracted from personal posted text, and wherein the selected variable subset of user features include personal interests.

### Claim 7 (depends on 5)

The method of claim 5, wherein each user feature profile is labelled with a label obtained from the selected target subset of the user features.

### Claim 8 (depends on 1)

The method of claim 1, wherein the user feature profiles are clustered using a k-means clustering model implementation of the clustering-component having a plurality of adjustable parameters that are set during the training of the clustering-component, and wherein the matching-component is implemented based on a regression model.

### Claim 9 (depends on 8)

The method of claim 8, wherein the plurality of adjustable parameters include wherein a number of clusters selected from the range 500-550 when a number of user features profiles is 40000-50000.

### Claim 10 (depends on 1)

The method of claim 1, wherein features extracted from structured user profiles are selected from the group consisting of: department, role, job profile, and geographic location.

### Claim 11 (depends on 1)

The method of claim 1, wherein features extracted from an analysis of at least one of:
unstructured user generated content posted on personal pages of the social network by users, members of users in special interest groups of the social network, and structured user profiles, are selected from the group consisting of: seniority, organization level distance, title, department, job description, author style, topic modelling, and expertise level and/or personal interest level.

### Claim 12 (depends on 1)

The method of claim 1, wherein at least one feature of the feature profile is computed from an analysis of membership in special interest groups of the social network is selected from the group consisting of: unstructured activity in each respective group, number of posts in each respective group, administrator status in groups, number of likes posted in each respective group, and number of group related keywords used in posts in the respective groups.

### Claim 13 (depends on 1)

The method of claim 1, wherein the feature profiles include a feature indicative of a seniority and expertise level of the respective user computed by a heuristic mapping of words extracted from an employee title associated with a profile of the respective user posted on the social network.

### Claim 14 (depends on 1)

The method of claim 1, wherein the extracting, the training the clustering-component, the training the matching-component, and the providing the model are iterated at spaced apart time intervals for dynamically updating the model using updated user features.

### Claim 15

A computer implemented method for establishing a communication session between an initiating client terminal and a target client terminal selected by a trained machine learning model, comprising:
receiving a request from an initiating client terminal of a certain user of a social network hosted by at least one social network server;
accessing a certain cluster computed by a clustering-component of the model, the certain cluster storing feature profiles of the certain user and other users of the social network, each user feature profile being structured and including a plurality of user features extracted from unstructured user generated text, indications of participation in groups, and structured user profiles;
computing, by a matching-component of the model, a distance score indicative of statistical similarity between a feature profile of the certain user and the features profiles of other users included in the certain cluster;
receiving, from the initiating client terminal, a selection of one of a subset of the other users; and
establishing the communication session between the initiating client, and the target client terminal of the selected one user.

### Claim 16 (depends on 15)

The method of claim 15, further comprising receiving from the initiating client terminal a target value of a certain user specific parameter, and wherein the statistical similarity is computed by matching user feature profiles of the cluster to the certain user specific parameter.

### Claim 17 (depends on 15)

The method of claim 15, wherein the distance score of each pair is computed as a statistical distance indicative of correlation score between the feature profiles of the users of the respective pair, or assigned a value denoting a maximal distance score provided by one of the users of the respective pair, or assigned a value denoting minimal distance score provided by one of the users of the respective pair.

### Claim 18 (depends on 17)

The method of claim 17, further comprising, after establishing the communication session, obtaining an indication of the distance score from the client terminal for update of the label of the respective pair, and updating the matching-component of the model.

### Claim 19 (depends on 18)

The method of claim 18, further comprising:
receiving from the initiating client terminal, an indication of no selection of the subset of the other users included in the certain cluster;
updating labels of pairs of the certain user and each of the subset of the other users with an indication of a maximal distance score;
updating the clustering-component and matching-component of the model with the updated labels of the pairs;
computing updated clusters by the updated clustering-component;
computing updated distance scores by the updated matching-component; and
receiving a new selection of a new subset of other users included in a new certain cluster presented on the initiating client terminal.

### Claim 20

A system for training a machine learning model for establishing a communication session, comprising:
at least one hardware processor executing a code for:
extracting a plurality of user feature profiles for a plurality of users of a social network hosted by at least one social network server, each user feature profile being structured and including a plurality of user features extracted from unstructured user generated text, indications of participation in groups, and structured user profiles;
training a clustering-component of the model to cluster the plurality of user feature profiles;
training a matching-component of the model to compute a distance score indicative of statistical similarity between a feature profile of a target user and features profiles of other users of a same cluster, using a training dataset of pairs of user feature profiles extracted from common clusters, each pair assigned a distance score label;
providing the model for:
identifying a certain cluster of a certain user; and
computing distance scores between the user feature profile of the certain user and other user feature profiles of other users of the certain cluster for selecting a subset of the other users;
wherein the communication session is established between an initiating client terminal of the certain user and the target client terminal of a user selected by the initiating client terminal from the subset.
