Automatic classification of messages based on keywords

Abstract

The invention discloses automatic classification of messages based on keywords. Electronic communications and keywords are provided to a machine learning algorithm. A similarity metric is received from a machine learning algorithm. The similarity metric indicates a similarity between the electronic communication and the keyword. And clustering the electronic communication according to the similarity measure. The false positron communications are removed from the first cluster according to a sum of distances between the false positron communications and the electronic communications including the keyword in the first cluster and a sum of distances between the false positron communications and the electronic communications not including the keyword in the first cluster. And adding the false negative electronic communication to the first cluster according to the sum of the distances between the false negative electronic communication and the electronic communication including the keyword in the first cluster and the sum of the distances between the false negative electronic communication and the electronic communication not including the keyword in the first cluster.

Description

Technical Field

[0001]

[0013] Implementations described herein generally relate to automatic categorization of messages based on keywords and, in implementations, but not by way of limitation, to automatic categorization of messages based on keywords using a machine learning algorithm. Background Art

[0002] Typical business personnel receive a large number of messages in the form of emails, chats and other electronic messages. Most of these people do not have the time to sort and organize these messages into appropriate categories, folders and / or files, such as organizing these communications according to various projects in the business. Conventional machine learning can be used to classify these messages, but conventional machine learning techniques need to create a dedicated model for each user to understand which messages belong to a specific project or category. This method requires a large amount of data, supervised learning from each user, and a large amount of local processing. For these reasons, conventional methods are not practical.

[0003] Currently, Long Short-Term Memory (LSTM) and Transformer neural networks are the main methods used to classify documents. However, similarly, these methods require a set of original sentences and their labels as training data. These methods may work well in classifying Internet articles because there is a large amount of data to identify these articles and the appropriate labels. However, it is impractical to prepare individual-specific classifiers because preparing a single classifier requires a large amount of training data and processing resources for training. Summary of the invention

[0004] According to one aspect of the present disclosure, a computerized process includes: providing electronic communications to a machine learning algorithm; providing keywords to the machine learning algorithm; receiving a similarity measure from the machine learning algorithm, the similarity measure indicating the similarity between the electronic communications and the keywords; clustering the electronic communications according to the similarity measure; removing the false positronic communications from the first cluster based on the sum of the distances between the false positronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false positronic communications and the electronic communications in the first cluster that do not include the keywords; and adding the false negative electronic communications to the first cluster based on the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that do not include the keywords.

[0005] According to another aspect of the present disclosure, a non-transitory machine-readable medium includes instructions that, when executed by a processor, perform a process, the process including: providing electronic communications to a machine learning algorithm; providing keywords to the machine learning algorithm; receiving a similarity measure from the machine learning algorithm, the similarity measure indicating the similarity between the electronic communications and the keywords; clustering the electronic communications according to the similarity measure; removing the false positron communication from the first cluster based on the sum of the distances between the false positron communication and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false positron communication and the electronic communications in the first cluster that do not include the keywords; and adding the false negative electronic communication to the first cluster based on the sum of the distances between the false negative electronic communication and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false negative electronic communication and the electronic communications in the first cluster that do not include the keywords.

[0006] According to another aspect of the present disclosure, a system includes: a computer processor; and a memory coupled to the computer processor; wherein the computer processor and the memory are operable to: provide electronic communications to a machine learning algorithm; provide keywords to the machine learning algorithm; receive a similarity measure from the machine learning algorithm, the similarity measure indicating the similarity between the electronic communications and the keywords; cluster the electronic communications according to the similarity measure; remove the false positronic communications from the first cluster based on the sum of the distances between the false positronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false positronic communications and the electronic communications in the first cluster that do not include the keywords; and add the false negative electronic communications to the first cluster based on the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that do not include the keywords. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In the accompanying drawings (which are not necessarily drawn to scale), similar reference numerals may describe similar components in different views. Similar reference numerals with different letter suffixes may represent different instances of similar components. In the various figures of the accompanying drawings, some embodiments are shown by way of example and not limitation.

[0008] Figure 1 The results of a sentence BERT (Bidirectional Encoder Representations from Transformers) operation are shown.

[0009] Figure 2 It is a Venn-like diagram of the results of a keyword search for electronic communications.

[0010] Figure 3 is a block diagram illustrating the operation and features of a system and process for classifying electronic communications.

[0011] Figure 4 Clustering of electronic communications is shown.

[0012] Figure 5 The calculation of the probability of classification of electronic communications is shown.

[0013] Figure 6 The calculation of the probability of classification of electronic communications is shown.

[0014] Figure 7 is a block diagram of a computer architecture upon which one or more of the embodiments disclosed herein may be executed. DETAILED DESCRIPTION

[0015] Based on the above difficulties in classifying electronic communications, there is a need to automatically classify messages into projects or other categories based on their natural language content. Implementations combine keyword searches and machine learning frameworks such as sentence BERT (Bidirectional Encoder Representations from Transformers) that can understand the meaning of ambiguous language to perform project-specific (and other) classification for company projects without creating training data for each user. Figure 1 An example result of a Sentence BERT operation is shown. Sentence 110 and sentence 120 are similar and therefore receive a higher similarity score. However, sentence 130 is not similar to sentence 110 or sentence 120 and therefore receives a lower similarity score.

[0016] Based on a set of project keywords, keyword search is applied to the BERT class model. A sentence BERT model can be used that converts sentences into vectors. The distance between these vectors is used to determine the degree of relevance between various sentences. In other words, if the correlation between two sentences is high, the two vectors will be close. If the correlation between two sentences is low, the two vectors will be far apart. Exploiting this feature, targeted classification is achieved using the trained model and inference.

[0017] Embodiments can pick up relevant electronic communications by combining sentence BERT and keyword searches. This process enables individual-specific classification and local processing that was previously impossible. In addition, embodiments can exclude irrelevant electronic communications by combining sentence BERT, which can quantify sentences, with clustering, and the process detects sentences that are less relevant to the project.

[0018] The relevance of electronic communications is calculated in a unique way. First, the user specifies at least one keyword related to a specific project (or other topic). The algorithm looks at all user messages (i.e. emails, chats, etc.) and classifies them into different parts. Figure 2 As shown, a Venn diagram of emails (or other electronic communications) detected by keyword search and emails related to the project is created. The project portion 220 of the Venn diagram contains: spelling errors, other relevant keywords for the project, and other emails that are related but do not include the project keywords. The keyword search can also identify messages that may contain keywords but are not related to the project, and are represented by portion 210. Based on the characteristics of each area, false positives in portion 210 are detected by X-means, and false negatives in portion 220 are detected by conditional probabilities represented by Gaussian functions. Portion 220A represents false negatives that should have been classified as positive, and portion 210A represents false positives that should have been classified as negative.

[0019] The algorithm works as follows. A keyword search is performed for the entire electronic communication using a registered word such as a project name. Since Sentence BERT creates a similarity measure between two sentences, it is expected that the extraction results will contain errors. The reason for this is that the second sentence is just a single project keyword rather than a sentence.

[0020] As above Figure 2 As discussed in the discussion of , there are two types of errors. One type of error is a false negative in a keyword search. A false negative is an electronic communication that is relevant to the project but was not detected by the keyword search. The keyword may not have been included, for example, because there was a spelling error of the keyword in the electronic communication, because a different project name was used, and because the project may have been discussed without using the project name keyword.

[0021] Another type of error is a false positive. These false positives are electronic communications that are detected by keyword searches, but they are not relevant to the project. Keywords may appear in electronic communications for one or more of the following reasons. For example, a project name may have been included in an electronic communication, but the project is not the main topic of the communication (i.e., the project is mentioned only because it is related to another project). As another example, a project name may appear in an electronic communication, but it is not relevant to the discussion (e.g., a social event where only the project team is discussed).

[0022] Detect false positive outliers from electronic communications by keyword search to detect false positives with low relevance. By doing so, false positives with low relevance can be reduced in subsequent steps. X-means process can be used for outlier detection. For false negatives, neighboring data detection is used. Use data based on which outliers are excluded from keyword detection data. Data near these data will be considered relevant to the project.

[0023] The following process is used to calculate the probability based on the proximity. As an initial value, the false positive data is set to a probability of 0.95, and all other data is set to a probability of 0.05. The reason for not setting it completely to 0 is to provide a buffer to prevent mislabeling caused by label smoothing.

[0024] Figure 3 is a more detailed block diagram illustrating an example implementation of a system and process for automatically categorizing messages based on keywords. Figure 3 Includes a plurality of process and feature blocks 310 to 360. Although Figure 3 In the examples of FIG. 1 , the blocks are arranged substantially serially, but other examples may reorder the blocks, omit one or more blocks, and / or execute two or more blocks in parallel using multiple processors or a single processor organized as two or more virtual machines or sub-processors.

[0025] Now refer to Figure 3 , at 310, an electronic communication is provided to a machine learning algorithm. The electronic communication may be an email, chat, text, or some other form of electronic communication 311. As indicated at 312, the machine learning algorithm may be an untrained machine learning algorithm. As further indicated at 313, the machine learning algorithm may be a Bidirectional Encoder Representations Based on Transformers (BERT) model. Such an untrained machine learning algorithm may understand ambiguous language 314.

[0026] At 320, the keywords are provided to the machine learning algorithm. As indicated at 322, a portion of the electronic communication provided to the machine learning algorithm can be stored in a first vector, and the keywords provided to the machine learning algorithm can be stored in a second vector.

[0027] At 330, a similarity metric is received from the machine learning algorithm. The similarity metric indicates a similarity between the electronic communication and the keyword.

[0028] At 340, the electronic communications are clustered according to a similarity measure. Figure 4 , where electronic communications 410A and 410B include the keyword. It is desirable to now include electronic communications that are false negatives, such as electronic communications that are relevant to the project but were not detected by the keyword search. The keyword may not have been included, for example, because there are misspellings of the keyword in the electronic communication, because a different project name is used, and because the project may have been discussed without using the project name keyword, such as Figure 4 As indicated by 420A, 420B and 420C.

[0029] At 350, the false positronic communications are removed from the first cluster based on the sum of the distances between the false positronic communications and the electronic communications in the first cluster that include the keyword and the sum of the distances between the false positronic communications and the electronic communications in the first cluster that do not include the keyword. Figure 5 is shown as the distance between communication 510 and communication 520. Figure 5 is further represented in equations 530 and 532. This process can also use the X-mean process.

[0030] At 360, the pseudo-negative electronic communication is added to the first cluster according to the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster that includes the keyword and the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster that does not include the keyword. Figure 6 The probability P of associating these distances and electronic communications with a given item is shown as the distance between communications 610 and 620. i Multiply, as indicated in equation 632. In an embodiment, the probability P i is a conditional probability represented by a Gaussian function. This method enables email classification without the need for personal data for training. P is updated by a specific process i , to obtain the probability that each email is associated with the item. As indicated in equation 630, by default, for electronic communications found through keyword searches, P i is set to 1, while for other electronic communications, P i The value of is set to 0. In an embodiment, values ​​0.95 and 0.05 are used for label smoothing. It is then updated according to equation 632 to increase the probability of each term that is not captured by the keyword search but is close in context.

[0031] Figure 7is a block diagram illustrating a computing and communications platform 700 in the example form of a general purpose machine according to various embodiments, wherein Figure 3 Some or all of the operations in can be performed on the computing and communication platform 700. In some embodiments, programming the computing platform 700 according to one or more specific algorithms produces a special-purpose machine when the programming is executed. In a networked deployment, the computing platform 700 can operate as a server or a client machine in a server-client network environment, or the computing platform 700 can act as a peer machine in a peer-to-peer (or distributed) network environment.

[0032] The example computing platform 700 includes at least one processor 702 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), or both, a processor core, a computing node, etc.), a main memory 701, and a static memory 706, which communicate with each other via a link 708 (e.g., a bus). The computing platform 700 may also include a video display unit 710, an input device 717 (e.g., a keyboard, a camera device, a microphone), and a user interface (UI) navigation device 711 (e.g., a mouse, a touch screen). The computing platform 700 may additionally include a storage device 716 (e.g., a drive unit), a signal generating device 718 (e.g., a speaker), a sensor 724, and a network interface device 720 coupled to a network 726.

[0033] The storage device 716 includes a non-transitory machine-readable medium 722 having stored thereon one or more sets of data structures and instructions 723 (e.g., software) embodied or embodied by any one or more of the methods or functions described herein. During execution of the instructions 723 by the computing platform 700, the instructions 723 may reside in whole or at least in part in the main memory 701, the static memory 706, and / or in the processor 702, where the main memory 701, the static memory 706, and the processor 702 also constitute machine-readable media.

[0034] Although the machine-readable medium 722 is shown as a single medium in the example embodiment, the term "machine-readable medium" may include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) storing one or more instructions 723. The term "machine-readable medium" should also be deemed to include any tangible medium that is capable of storing, encoding, or carrying instructions that are executed by a machine and cause the machine to perform any one or more of the methods of the present disclosure, or that is capable of storing, encoding, or carrying data structures utilized by or associated with such instructions. Therefore, the term "machine-readable medium" should be deemed to include, but is not limited to, solid-state memory and optical and magnetic media. Specific examples of machine-readable media include non-volatile memory, which includes, by way of example, but is not limited to, semiconductor memory devices (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)) and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.

[0035] The above detailed description includes reference to the accompanying drawings, which form a part of the detailed description. The accompanying drawings show the specific embodiments that can be put into practice by way of illustration. These embodiments are also referred to as "examples" herein. Such examples may include elements other than the elements shown or described. However, examples including the elements shown or described are also contemplated. In addition, examples (or one or more aspects of examples) using any combination or arrangement of those elements shown or described are also contemplated for a specific example (or one or more aspects of a specific example) or for other examples (or one or more aspects of other examples) shown or described herein.

[0036] Publications, patents, and patent documents cited in this document are incorporated herein by reference in their entirety, as if individually incorporated by reference. In the event of an inconsistency between the usages of this document and those documents so incorporated by reference, the usage in the incorporated reference(s) supplements that of this document; for conflicting inconsistencies, the usage in this document controls.

[0037] In this document, as is common in patent documents, the terms "a" or "an" are used to include one or more than one, independent of any other instance or usage of "at least one" or "one or more". In this document, unless otherwise indicated, the term "or" is used to refer to a non-exclusive or, such that "A or B" includes "A but not B", "B but not A", and "A and B". In the appended claims, the terms "including" and "in which" are used as the plain English equivalents of the corresponding terms "comprising" and "wherein". In addition, in the appended claims, the terms "including" and "comprising" are open-ended, that is, systems, devices, articles, or processes that include elements other than those listed after such terms in the claim are still considered to fall within the scope of the claim. In addition, in the appended claims, the terms "first", "second", and "third", etc. are used only as labels and are not intended to imply a numerical order of their objects.

[0038] The above description is intended to be illustrative, not restrictive. For example, the above examples (or one or more aspects of the above examples) may be used in combination with other examples. For example, a person of ordinary skill in the art may use other embodiments after reviewing the above description. The abstract enables the reader to quickly determine the nature of the technical disclosure. The abstract is submitted with the following understanding: the abstract will not be used to interpret or limit the scope or meaning of the claims. In addition, in the above specific embodiments, various features may be grouped together to organize the present disclosure. However, since the embodiments may be characterized by a subset of the features, the claims may not set forth each feature disclosed herein. In addition, the embodiments may include fewer features than those disclosed in a particular example. Therefore, the attached claims are hereby incorporated into the specific embodiments, wherein the claims themselves are separate embodiments. The scope of the embodiments disclosed herein is to be determined with reference to the attached claims and the full scope of equivalents given by these claims.

[0039] Example

[0040] Example 1 is a process comprising: providing electronic communications to a machine learning algorithm; providing keywords to the machine learning algorithm; receiving a similarity measure from the machine learning algorithm, the similarity measure indicating the similarity between the electronic communications and the keywords; clustering the electronic communications according to the similarity measure; removing false positive electronic communications from a first cluster based on the sum of the distances between the false positive electronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false positive electronic communications and the electronic communications in the first cluster that do not include the keywords; and adding false negative electronic communications to the first cluster based on the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that do not include the keywords.

[0041] Example 2 includes all the features of Example 1, and optionally includes a process wherein the machine learning algorithm is an untrained machine learning algorithm.

[0042] Example 3 includes all of the features of Examples 1-2, and optionally includes a process wherein the machine learning algorithm understands ambiguous language.

[0043] Example 4 includes all of the features of Examples 1 to 3, and optionally includes a process wherein the machine learning algorithm comprises a Bidirectional Encoder Representation based Transformer (BERT) model.

[0044] Example 5 includes all of the features of Examples 1 to 4, and optionally includes a process comprising storing a portion of an electronic communication provided to the machine learning algorithm in a first vector and storing keywords provided to the machine learning algorithm in a second vector.

[0045] Example 6 includes all of the features of Examples 1 to 5, and optionally includes the process wherein the electronic communication includes one or more of email, chat, and text.

[0046] Example 7 is a machine-readable medium, which includes instructions that, when executed by a processor, perform a process, the process including: providing electronic communications to a machine learning algorithm; providing keywords to the machine learning algorithm; receiving a similarity measure from the machine learning algorithm, the similarity measure indicating the similarity between the electronic communications and the keywords; clustering the electronic communications according to the similarity measure; removing false positive electron communications from the first cluster based on the sum of the distances between the false positive electron communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false positive electron communications and the electronic communications in the first cluster that do not include the keywords; and adding false negative electron communications to the first cluster based on the sum of the distances between the false negative electron communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false negative electron communications and the electronic communications in the first cluster that do not include the keywords.

[0047] Example 8 includes all the features of Example 7, and optionally includes a machine-readable medium wherein the machine learning algorithm is an untrained machine learning algorithm.

[0048] Example 9 includes all of the features of Examples 7 to 8, and optionally includes a machine-readable medium wherein the machine learning algorithm understands ambiguous language.

[0049] Example 10 includes all of the features of Examples 7 to 9, and optionally includes a machine-readable medium wherein the machine learning algorithm comprises a Bidirectional Encoder Representation based Transformer (BERT) model.

[0050] Example 11 includes all of the features of Examples 7 to 10, and optionally includes a machine-readable medium comprising instructions for storing a portion of an electronic communication provided to a machine learning algorithm in a first vector and storing keywords provided to the machine learning algorithm in a second vector.

[0051] Example 12 includes all of the features of Examples 7 to 11, and optionally includes the machine-readable medium: wherein the electronic communication includes one or more of email, chat, and text.

[0052] Example 13 is a system comprising: a computer processor; and a memory coupled to the computer processor; wherein the computer processor and the memory are operable to: provide electronic communications to a machine learning algorithm; provide keywords to the machine learning algorithm; receive a similarity measure from the machine learning algorithm, the similarity measure indicating the similarity between the electronic communications and the keywords; cluster the electronic communications according to the similarity measure; remove false positive electronic communications from a first cluster based on the sum of the distances between the false positive electronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false positive electronic communications and the electronic communications in the first cluster that do not include the keywords; and add false negative electronic communications to the first cluster based on the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that include the keywords and the sum of the distances between the false negative electronic communications and the electronic communications in the first cluster that do not include the keywords.

[0053] Example 14 includes all of the features of Example 13, and optionally includes a system wherein the machine learning algorithm is an untrained machine learning algorithm.

[0054] Example 15 includes all of the features of Examples 13-14, and optionally includes a system wherein the machine learning algorithm understands ambiguous language.

[0055] Example 16 includes all of the features of Examples 13 to 15, and optionally includes a system wherein the machine learning algorithm comprises a Bidirectional Encoder Representation based Transformer (BERT) model.

[0056] Example 17 includes all of the features of Examples 13 to 16, and optionally includes a system operable to store a portion of an electronic communication provided to a machine learning algorithm in a first vector, and instructions to store keywords provided to the machine learning algorithm in a second vector.

[0057] Example 18 includes all of the features of Examples 13 to 17, and optionally includes the system wherein the electronic communication includes one or more of email, chat, and text.

Claims

1. A computerized process comprising: providing electronic communications to machine learning algorithms; providing keywords to the machine learning algorithm; receiving a similarity measure from the machine learning algorithm, the similarity measure indicating a similarity between the electronic communication and the keyword; clustering the electronic communications according to the similarity measure; removing the false positron communication from the first cluster according to the sum of the distances between the false positron communication and the electronic communications in the first cluster that include the keyword and the sum of the distances between the false positron communication and the electronic communications in the first cluster that do not include the keyword; as well as The pseudo-negative electronic communication is added to the first cluster according to the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster including the keyword and the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster not including the keyword.

2. The computerized process of claim 1, wherein: The machine learning algorithm is an untrained machine learning algorithm.

3. The computerized process of claim 1, wherein: The machine learning algorithm understands ambiguous language.

4. The computerized process of claim 1, wherein: The machine learning algorithm includes a Transformer-based bidirectional encoder representation BERT model.

5. The computerized process of claim 1, comprising: A portion of the electronic communication provided to the machine learning algorithm is stored in a first vector, and keywords provided to the machine learning algorithm are stored in a second vector.

6. The computerized process of claim 1, wherein: The electronic communication includes one or more of email, chat, and text.

7. A non-transitory machine-readable medium comprising instructions that, when executed by a processor, perform a process comprising: providing electronic communications to machine learning algorithms; providing keywords to the machine learning algorithm; receiving a similarity metric from the machine learning algorithm, the similarity metric indicating a similarity between the electronic communication and the keyword; clustering the electronic communications according to the similarity measure; removing the false positron communication from the first cluster according to the sum of the distances between the false positron communication and the electronic communications in the first cluster that include the keyword and the sum of the distances between the false positron communication and the electronic communications in the first cluster that do not include the keyword; as well as The pseudo-negative electronic communication is added to the first cluster according to the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster including the keyword and the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster not including the keyword.

8. The non-transitory machine-readable medium of claim 7, wherein: The machine learning algorithm is an untrained machine learning algorithm.

9. The non-transitory machine-readable medium of claim 7, wherein: The machine learning algorithm understands ambiguous language.

10. The non-transitory machine-readable medium of claim 7, wherein: The machine learning algorithm includes a Transformer-based bidirectional encoder representation BERT model.

11. The non-transitory machine-readable medium of claim 7, comprising instructions for storing a portion of an electronic communication provided to the machine learning algorithm in a first vector and storing keywords provided to the machine learning algorithm in a second vector.

12. The non-transitory machine-readable medium of claim 7, wherein: The electronic communication includes one or more of email, chat, and text.

13. A system comprising: Computer processors; as well as a memory coupled to the computer processor; Wherein, the computer processor and the memory are operable to: providing electronic communications to machine learning algorithms; providing keywords to the machine learning algorithm; receiving a similarity metric from the machine learning algorithm, the similarity metric indicating a similarity between the electronic communication and the keyword; clustering the electronic communications according to the similarity measure; removing the false positron communication from the first cluster according to the sum of the distances between the false positron communication and the electronic communications in the first cluster that include the keyword and the sum of the distances between the false positron communication and the electronic communications in the first cluster that do not include the keyword; and The pseudo-negative electronic communication is added to the first cluster according to the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster including the keyword and the sum of the distances between the pseudo-negative electronic communication and the electronic communication in the first cluster not including the keyword.

14. The system according to claim 13, wherein: The machine learning algorithm is an untrained machine learning algorithm.

15. The system of claim 13, wherein: The machine learning algorithm understands ambiguous language.

16. The system of claim 13, wherein: The machine learning algorithm includes a Transformer-based bidirectional encoder representation BERT model.

17. The system of claim 13, wherein: The system can be operable to store a portion of an electronic communication provided to the machine learning algorithm in a first vector and store keywords provided to the machine learning algorithm in a second vector.

18. The system of claim 13, wherein: The electronic communication includes one or more of email, chat, and text.

Images