Browse Prior Art Database

Method for suggesting recipients to text messages based on content Disclosure Number: IPCOM000223811D
Publication Date: 2012-Nov-29
Document File: 2 page(s) / 220K

Publishing Venue

The Prior Art Database


Mobile device users must pass several hurdles when sending text messages to other users. Recipients must be selected from a list, and there is a possibility that the wrong recipient may be selected. This article describes a method for overcoming these two issues by using a machine learning approach to predict text message recipients based on natural language processing of the message content. Warnings can be issued if attempting to send a message to an incorrect recipient, and recipient selection can be eliminated altogether by typing the message first and selecting from a likely list of recipients. As more messages are sent to a user's contacts, the system will train itself to better predict the recipient. The system includes pre-trained models to work "out of the box" to predict groups of contacts, and it will become more accurate over time as the models are trained for individual contacts.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 2

Method for suggesting recipients to text messages based on content


* A mobile device user accidentally sends a text message to the wrong recipient.

* The mobile device user may have difficulty or take significant time locating a single contact within a large contact list.


Several documents that describe the concept of predicting message recipients were discovered. None of them described the use of independent scorers and classifiers to narrow down the recipient list. Some solutions require accruing message history to be able to start using the functionality, other worked out of a set of defined patterns. These solutions are felt to be restrictive and do not provide the potential of looking at a variety of aspects about the messages (Content, Time, Message Count, Length, etc). Additionally, the proposed system has the ability to learn based on the history of message and build new specific classifiers for the most important recipients. - Different mechanics and requires a history to be able to work - Different mechanics, requires pattern definitions and - Add additional message recipients based on message analysis, requires an initial recipient

User Interface Use Cases:

1. Create a new message

* User types the text up front. The pattern is reversed from the convention of first selecting a recipient and then typing text

* User selects the recipient from a reduces set of contact that is determined by the system based on running the classification. The system pre-fills with the most likely recipient.

Main advantage is a reduction of friction when it comes to typing a message. Finding the recipient first is often an impediment, specially with large contact lists.

2. Avoid sending messages to Unintended recipient

* User starts typing a message within a conversation screen, but the user does not notice that the wrong recipient conversation is in context

* After submitting the message, the system will classify the message and compare with the group of the current recipient.

* If there is no match, the system will alert the user to confirm the recipient.

Main advantage is that the messaging system will provide a validation step before sending messages to make sure that they are going to the intended recipient and allows user intervention.

Recipient Suggestion:

1. Text message is run through a set of scores

2. The score vector is then ran through a set of independent classifiers.

3. Each classifier will provide the probability of the text ta...