Filtering electronic messages

• US 10,447,635 B2
• Filed: 05/17/2017
• Issued: 10/15/2019
• Est. Priority Date: 05/17/2017
• Status: Active Grant

• Current Assignee: Nielsen Consumer LLC (Nielsen Holdings plc)
• Original Assignee: Slice Technologies, Inc. (Rakuten Group, Inc.)
• Inventors: Sathi, Conal, Tarasov, Alexander, Mykhaylov, Dmytro, Kokhlikyan, Narine, Ivchenko, Roman
• Primary Examiner(s): Al-Hashemi, Sana A
• Application Number:
  US15/597,404
• Publication Number:
  US 20180337878A1
• Time in Patent Office: 881 Days
• Field of Search: 707687
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]
• CPC Class Codes: H04L 51/212 using filtering or selective blocking

Abstract

Improved systems and methods for automatically discovering and filtering electronic messages. These systems and methods improve the operation of computer apparatus to achieve dramatic reductions in processing resources, data storage resources, network resources, and filter production times compared to conventional approaches. In some examples, improvements result from configuring computer apparatus to perform a unique sequence of specific electronic message processing rules in a network communications environment. In this regard, these examples are able to automatically learn the structures and semantics of machine generated electronic message headers, accelerating the ability to support new message sources and new markets. These examples provide a purchase related electronic message discovery and filtering service that is able to identify and filter purchase related electronic messages with high accuracy across a wide variety of electronic message formats.

Claims

1. A method for filtering a population of electronic messages transmitted between network nodes and stored in association with respective user accounts on one or more network data storage systems managed by one or more messaging servers, each electronic message being associated with a respective sender, a respective header, and a respective body, the method comprising,

by a client network node fetching, from one or more of the messaging servers, headers in the population that are stored on one or more of the network data storage systems across multiple of the user accounts;

for each of one or more senders, grouping the fetched headers that are associated with the sender into clusters based on a density based clustering process that correlates the clusters with respective dense regions in a clustering data space in which the fetched headers are separated from one another based on similarities between respective pairs of the fetched headers;

for each of one or more of the clusters,

by a client network node retrieving, from one or more of the messaging servers, a respective sample of the electronic messages in the population that are associated with the fetched headers in the cluster and stored on one or more of the network data storage systems;

classifying, with a machine learning classifier, each electronic message in the retrieved sample with a respective label from a predefined set of labels comprising one or more purchase related labels and an associated confidence level to create a respective classification data set for the cluster;

assigning to the cluster a respective cluster label selected from the predefined set of labels based on at least one cluster classification rule that maps the respective classification data set to the respective cluster label;

for each of one or more clusters assigned a respective one of the purchase related labels, automatically generating a respective filter for filtering purchase related electronic messages; and

installing, by a processor, one or more of the filters in at least one network communication channel to select purchase related electronic messages from a set of electronic messages stored in association with respective user accounts on one or more network data storage systems managed by one or more messaging servers.

2. The method of claim 1, wherein the grouping comprises calculating similarity scores between respective pairs of the fetched messages headers based on comparisons of strings in the subject fields of the respective message headers, and grouping the fetched headers into clusters based on the calculated similarity scores.

3. The method of claim 2, wherein the grouping comprises applying a similarity threshold to the calculated similarity scores to generate the clusters of the fetched message headers.

4. The method of claim 3, wherein the grouping comprises iteratively adjusting the similarity threshold and grouping the fetched message headers into clusters based on the adjusted similarity threshold until a count of the clusters is determined to have converged.

5. The method of claim 4, wherein the adjusting comprises adjusting the similarity threshold progressively lower from an initial similarity threshold value.

6. The method of claim 4, wherein, for at least one iteration of the adjusting, the grouping comprises requiring that the headers in each cluster have words in common.

7. The method of claim 1, wherein the grouping is performed without regard to the bodies associated with the headers in the sample.

8. The method of claim 1, wherein each sample of the retrieved electronic messages consists of a predetermined number of electronic messages.

9. The method of claim 1, wherein the classifying of each electronic message is based on contents of the header and the body that are respectively associated with the electronic message.

10. The method of claim 1, wherein the predefined set of labels comprises a first label to categorize an electronic message as a receipt and a second label to categorize an electronic message as not a receipt.

11. The method of claim 1, wherein the generating of a respective filter comprises identifying common substrings within the headers in the respective cluster, and generating a respective filter based on the identified substrings.

12. The method of claim 11, wherein the generating of a respective filter comprises determining respective counts of bigrams appearing in the headers in the respective cluster, and incorporating one or more of the bigrams associated with prominent counts into the respective filter rule.

13. The method of claim 11, wherein each of respective ones of the filter rules defines a matching pattern for a respective set of subject field strings in an electronic message.

14. The method of claim 1, wherein the installing comprises installing one or more of the filters between a client network node and one or more of the messaging servers.

15. The method of claim 1, wherein each filter is associated with a respective set of one or more extraction parsers;

and

further comprising, for each filter matched to a respective one of the electronic messages in the set, attempting to parse the electronic message with the one or more extraction parsers associated with the matched filter.

16. Apparatus for generating filters operable to filter a population of electronic messages transmitted between network nodes and stored in association with respective user accounts on one or more network data storage systems managed by one or more messaging servers, each electronic message being associated with a respective sender, a respective header, and a respective body, the apparatus comprising a memory storing processor-readable instructions, and a processor coupled to the memory, operable to execute the instructions, and based at least in part on the execution of the instructions operable to perform operations comprising,

fetching headers in the population from one or more of the network data storage systems;

for each of one or more senders, grouping the fetched headers into clusters, wherein the grouping comprises assigning ones of the fetched headers to respective ones of the clusters based on similarities between the headers in the clusters without regard to any message body content;

for each of one or more of the clusters,

retrieving, from one or more of the network data storage systems, a respective sample of one or more of the electronic messages associated with the fetched headers assigned to the cluster, and

designating, with a machine learning classifier, the cluster as either receipt-related or not-receipt-related based on header and body content of the one or more retrieved electronic messages in the sample; and

automatically generating a respective electronic message filter for each of one or more of the clusters designated as receipt-related, wherein each electronic message filter defines a respective rule for matching a respective pattern of subject field strings in a header of an electronic message.

17. The method of claim 16, wherein the grouping comprises iteratively adjusting a similarity threshold and grouping the fetched message headers into clusters based on the adjusted similarity threshold until a count of the clusters is determined to have converged.

18. The apparatus of claim 16, wherein, based at least in part on the execution of the instructions, the processor is operable to perform operations further comprising:

installing one or more of the filters in at least one network communication channel to select purchase related electronic messages from a set of electronic messages stored in association with respective user accounts on one or more network data storage systems managed by one or more messaging servers.

19. The method of claim 16, wherein the fetching comprises fetching a sample of the headers in the population that are associated with the sender and stored on one or more of the network data storage systems across multiple of the user accounts.

20. At least one computer-readable medium having computer-readable program code embodied therein, the computer-readable program code adapted to be executed by a client network node to implement a method for processing a population of electronic messages transmitted between network nodes and stored in association with respective user accounts on one or more network data storage systems managed by one or more messaging servers, each electronic message being associated with a respective sender, a respective header, and a respective body, the method comprising,

by a client network node fetching, from one or more of the messaging servers, headers in the population that are stored on one or more of the network data storage systems across multiple of the user accounts;

for each of one or more senders, grouping the fetched headers that are associated with the sender into clusters based on a density based clustering process that correlates the clusters with respective dense regions in a clustering data space in which the fetched headers are separated from one another based on similarities between respective pairs of the fetched headers;

for each of one or more of the clusters,

by a client network node retrieving, from one or more of the messaging servers, a respective sample of the electronic messages in the population that are associated with the fetched headers in the cluster and stored on one or more of the network data storage systems;

classifying, with a machine learning classifier, each electronic message in the retrieved sample with a respective label from a predefined set of labels comprising one or more purchase related labels and an associated confidence level to create a respective classification data set for the cluster;

assigning to the cluster a respective cluster label selected from the predefined set of labels based on at least one cluster classification rule that maps the respective classification data set to the respective cluster label; and

for each of one or more clusters assigned a respective one of the purchase related labels, automatically generating a respective filter for filtering purchase related electronic messages.