Automated message attachment labeling using feature selection in message content
View Patent ↗Embodiments are directed towards an automated machine learning framework to extract keywords within a message that are relevant to an attachment to the message. The machine learning model finds a set of relevant sentences within the message determined to be relevant to the one or more attachments based on identification of one or more sentence level features within a given sentence. The sentence level features include, for example, anchor features, noisy sentence features, short message features, threading features, anaphora detections, and lexicon features. From the set of relevant sentences, useful keywords may be extracted using a sequence of heuristics to convert the sentence set into the set of useful keywords. The set of useful keywords may then be associated to at least one attachment such that the keywords may subsequently be used to perform various indexing, searching, sorting, and to provide further context to the attachment.
1. A network device, comprising:
a transceiver to send and receive data over a network; and
a processor that is operative on the received data to perform actions, including:
receiving a message having at least one attachment;
using a machine learning model to select at least one sentence from within the message to be relevant to the at least one attachment based on a set of predefined sentence level features;
identifying from within the relevant sentence at least one keyword determined to be further relevant to the at least one attachment; and
associating the at least one keyword to the at least one attachment, such that the association is useable for at least one of indexing or searching of the at least one attachment.
2. The network device of claim 1 , wherein the predefined sentence level features comprises at least one of an anchor feature, a noise feature, a length feature, a thread level feature, an anaphora detection feature, or a lexicon feature.
3. The network device of claim 1 , wherein the at least one keyword is identified based on employing a sequence of heuristics, including:
replacing a relative reference with an explicit reference within the relevant sentence;
replacing a pronoun with a noun for which the pronoun refers within the relevant sentence;
removing information determined to be sensitive or objectionable from within the relevant sentence; and
selectively removing stop words within the relevant sentence, such that at least one resulting word is identified as the at least one keyword.
4. The network device of claim 1 , wherein the message includes at least one of an email message, SMS message, MMS message, or IM message.
5. The network device of claim 1 , wherein the at least one attachment includes at least one of a spreadsheet file, an image file, a video file, an audio file, or a word processing document file.
6. The network device of claim 1 , wherein the predefined sentence level features include at least one of an anchor feature selected from strong phrase anchors, extension anchors, attachment name anchors, or weak phrase anchors.
7. The network device of claim 1 , wherein the processor is operative to provide for display a user interface useable to enable a review or deletion of at least one keyword.
8. A system, comprising:
a network device comprising a first processor and configured to manage a messaging component for receiving and sending messages over a network; and
a tagging component residing on another network device comprising a second processor, the tagging component operative to perform actions, including:
receiving a message having at least one attachment from the messaging component;
using a machine learning model to select at least one sentence from within the message to be relevant to the at least one attachment based on a set of predefined sentence level features;
identifying from within the at least one relevant sentences a set of keywords further relevant to the at least one attachment; and
associating at least one of the keywords in the set to the at least one attachment, such that the association is useable for at least one of indexing or searching of the at least one attachment.
9. The system of claim 8 , wherein the predefined sentence level features comprises at least one of an anchor feature, a noise feature, a length feature, a thread level feature, an anaphora detection feature, or a lexicon feature.
10. The system of claim 8 , wherein the other network device is a client device, and the tagging component is a downloadable component onto the client device.
11. The system of claim 8 , wherein the set of keywords are identified based on employing a sequence of heuristics, including:
replacing a relative reference with an explicit reference within the at least one relevant sentence;
replacing a pronoun with a noun for which the pronoun refers within the at least one relevant sentence;
removing information determined to be sensitive or objectionable from within the at least one relevant sentence; and
selectively removing stop words within the at least one relevant sentence, such that the resulting words are identified as the set of keywords.
12. The system of claim 8 , wherein the tagging component is configured to employ feedback about the set of keywords and to learn over time to predict at least one other keyword that is to be retained or deleted from the set of keywords.
13. The system of claim 8 , wherein the machine learning model includes a classification learning model.
14. A non-transitory computer-readable storage medium having computer-executable instructions, the computer-executable instructions when installed onto a computing device enable the computing device to perform actions, comprising:
receiving a message having at least one attachment;
using a machine learning model to select a set of sentences from within the message to be relevant to the at least one attachment based on a set of predefined sentence level features;
identifying from within the set of relevant sentences a set of keywords further relevant to the at least one attachment; and
associating at least one of the keywords in the set to the at least one attachment, such that the association is useable for at least one of indexing or searching of the at least one attachment.
15. The non-transitory computer-readable storage medium of claim 14 , wherein if the message includes a plurality of attachments, then associating the at least one of the keywords in the set to less that all of the attachments in the plurality of attachments.
16. The non-transitory computer-readable storage medium of claim 14 , wherein the predefined sentence level features comprise at least one of an anchor feature, a noise feature, a length feature, a thread level feature, an anaphora detection feature, or a lexicon feature.
17. The non-transitory computer-readable storage-medium of claim 14 , wherein the set of keywords are identified based on employing a sequence of heuristics to modify the words with the set of relevant sentences.
18. The non-transitory computer-readable storage medium of claim 14 , wherein a user interface is provided that enables removal of at least one keyword from the set of keywords.
19. The non-transitory computer-readable storage medium of claim 14 , wherein the instructions further predict at least one keyword within the set of keywords that is to be removed based in part on a prior removal of at least one other keyword.
20. The non-transitory computer-readable storage medium of claim 14 , wherein the computer-executable instructions are configured to be downloadable and executed from within a client device, or configured to operate within a cloud architecture.