CN114244795B - Information pushing method, device, equipment and medium - Google Patents
Information pushing method, device, equipment and medium Download PDFInfo
- Publication number
- CN114244795B CN114244795B CN202111544848.7A CN202111544848A CN114244795B CN 114244795 B CN114244795 B CN 114244795B CN 202111544848 A CN202111544848 A CN 202111544848A CN 114244795 B CN114244795 B CN 114244795B
- Authority
- CN
- China
- Prior art keywords
- target
- information
- intention
- instant message
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 230000004044 response Effects 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 84
- 238000012545 processing Methods 0.000 claims description 29
- 238000005315 distribution function Methods 0.000 claims description 24
- 238000004891 communication Methods 0.000 claims description 20
- 238000002372 labelling Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 230000000873 masking effect Effects 0.000 claims description 7
- 239000013598 vector Substances 0.000 claims description 7
- 238000007635 classification algorithm Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 5
- 230000010365 information processing Effects 0.000 claims description 4
- 238000007670 refining Methods 0.000 claims 1
- 238000013473 artificial intelligence Methods 0.000 abstract description 5
- 238000013135 deep learning Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 11
- 230000008901 benefit Effects 0.000 description 10
- 238000004590 computer program Methods 0.000 description 10
- 230000003442 weekly effect Effects 0.000 description 10
- 230000008569 process Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000007477 logistic regression Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/04—Real-time or near real-time messaging, e.g. instant messaging [IM]
- H04L51/043—Real-time or near real-time messaging, e.g. instant messaging [IM] using or handling presence information
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/07—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail characterised by the inclusion of specific contents
- H04L51/10—Multimedia information
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Data Mining & Analysis (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
The disclosure provides a method, a device, equipment and a medium for pushing information, which relate to the technical field of artificial intelligence, in particular to the technical field of deep learning, and comprise the following steps: entity identification is carried out on the input first instant message, and a target entity is obtained; acquiring target intention information corresponding to the target entity; in response to determining that the first instant message matches the target intent message, obtaining information matching the target intent message; and pushing the information to the target terminal. According to the technical scheme, the intention in the input instant message can be effectively and accurately identified, so that the message matched with the intention can be accurately pushed, and the efficiency and accuracy of information pushing are improved.
Description
Technical Field
The disclosure relates to the technical field of artificial intelligence, in particular to the technical field of deep learning, and specifically relates to a method, a device, equipment and a medium for pushing information.
Background
In instant messaging sessions, terms that are not known to the user typically appear, which are generally widely distributed and include encyclopedia, abbreviations in office scenes, reference classes, business products and items, and business cultures. For the message hitting the vocabulary entry, because the knowledge backgrounds of different users are different, how to judge whether the users have intention of knowing and using the vocabulary entry can quickly break the communication barriers among the users, and the communication efficiency among the users is improved.
In the related art, the lack of an effective intention recognition method suitable for an instant messaging scene results in lower communication efficiency between chat users.
Disclosure of Invention
The disclosure provides a method, a device, equipment and a medium for pushing information.
According to an aspect of the present disclosure, there is provided a pushing method of information, the method including:
entity identification is carried out on the input first instant message, and a target entity is obtained;
acquiring target intention information corresponding to the target entity;
in response to determining that the first instant message matches the target intent message, obtaining information matching the target intent message;
and pushing the information to the target terminal.
According to another aspect of the present disclosure, there is provided an information pushing apparatus, including:
the entity acquisition module is used for carrying out entity identification on the input first instant message to acquire a target entity;
the intention acquisition module is used for acquiring target intention information corresponding to the target entity;
the information acquisition module is used for responding to the fact that the first instant information is matched with the target intention information, and acquiring information matched with the target intention information;
And the information pushing module is used for pushing the information to the target terminal.
According to another aspect of the present disclosure, there is provided an electronic device including:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of the embodiments of the present disclosure.
According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method of any one of the embodiments of the present disclosure.
According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements a method as described in any of the embodiments of the present disclosure.
According to the technical scheme, the intention in the input instant message can be effectively and accurately identified, so that the message matched with the intention can be accurately pushed, and the efficiency and accuracy of information pushing are improved.
It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.
Drawings
FIG. 1 is a schematic diagram of a method of pushing information according to an embodiment of the present disclosure;
FIG. 2 is a schematic diagram of a method of pushing information according to yet another embodiment of the present disclosure;
FIG. 3a is a schematic diagram of a method of pushing information according to yet another embodiment of the present disclosure;
FIG. 3b is a schematic diagram of a training method of an intent recognition model in accordance with an embodiment of the present disclosure;
FIG. 3c is a schematic diagram of the structure of an ERNIE model according to an embodiment of the disclosure;
fig. 4 is a schematic structural diagram of an information pushing device according to an embodiment of the present disclosure;
fig. 5 is a block diagram of an electronic device of a method of pushing information according to an embodiment of the present disclosure.
Detailed Description
Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
Fig. 1 is a flow chart of a method for pushing information according to an embodiment of the present disclosure, where the embodiment is suitable for a situation that instant information is intended to be identified in an instant messaging scenario and matched information is pushed to a terminal according to an intended identification result, the method may be performed by an information pushing device, and the device may be implemented by software and/or hardware, and may be generally integrated in a terminal or a server having a data processing function. Specifically, referring to fig. 1, the method specifically includes the following steps:
step 110, entity identification is performed on the input first instant message, and a target entity is obtained.
In this embodiment, the instant messaging (InstantMessaging, IM) scenario may be a scenario in which real-time communication is performed through a network, allowing two or more people to communicate with video by instant messaging of text messages, files, voice. In the instant messaging scenario, chat information (i.e., first instant information) input by each chat user may be respectively obtained in the corresponding chat application program.
The first instant message in this embodiment needs to be acquired under the authorization of the user, and the first instant message is not information acquired for a specific user, and cannot reflect personal information of a specific user. That is, the above-mentioned first instant message is only used as a basis for the subsequent intention recognition and the information push, and is not used for other purposes.
It should be emphasized that, in the technical solution of the present disclosure, the processes of collecting, storing, using, processing, transmitting, providing, and disclosing the related personal information of the user all conform to the rules of the related laws and regulations, and do not violate the public welfare.
In this step, after the first instant message is obtained, optionally, the target entity included in the first instant message may be obtained through a preset entity identification manner.
Optionally, the preset universal chat vocabulary can be screened out from the first instant message, information that the chat user may have use or know intention in the first instant message is reserved, and the information is used as a target entity.
In a specific embodiment, the universal chat vocabulary may be a vocabulary that is frequently present in instant messaging scenarios, such as "good," clear, "and" thank you. Optionally, according to the universal chat vocabulary, the words consistent with or similar to the universal chat vocabulary can be deleted in the first instant message, so as to obtain the target entity. Assuming that the first instant message acquired is "ask someone to know what the CTR is," the "CTR" may be taken as the target entity to be identified.
Alternatively, an entity library may be pre-established, and the target entity included in the first instant message may be identified by matching each word included in the first instant message with the entity library.
Accordingly, in this embodiment, before the first instant information input by each chat user is obtained, an entity library may be pre-established, where the entity library is used to store a plurality of professional entities, and relates to encyclopedias, abbreviations under office scenes, reference classes, enterprise products or projects, enterprise culture classes, and so on.
Step 120, obtaining target intention information corresponding to the target entity.
In this step, a correspondence between the entity and the intention information may be established in advance, and then, after the target entity in the first instant message is acquired, the matched target intention information may be directly determined.
The target intention information refers to an actual requirement for a target entity, for example, if the target entity is a specific vocabulary, the target intention information may be a paraphrase acquisition requirement for the specific vocabulary, or if the target entity is a colleague of a chat user, the target intention information may be a contact acquisition requirement for the colleague, that is, a person finding requirement.
Alternatively, there may be a one-to-one correspondence between the entity and the intention information, or there may be a correspondence to many, which is not limited herein. For example, if the entity is a specific name, the matching intent information may be a paraphrase requirement for the name, or a contact acquisition requirement for the name.
Step 130, in response to determining that the first instant message matches the target intent message, obtaining information matching the target intent message.
In an optional implementation manner of this embodiment, after obtaining the target intention information, it may be detected whether the first instant message includes an action vocabulary that represents an obvious intention, for example, if the first instant message includes a vocabulary associated with the paraphrasing requirement such as "know", "know" and the like, for the paraphrasing requirement of the user, it may be determined that the first instant message matches the target intention information; for another example, for the user's find demand, if a word associated with the paraphrase demand, such as "contact" or "find" is included in the first instant message, it may be determined that the first instant message matches the target intent information.
In another optional implementation manner of this embodiment, the first instant message and the matched target intention information may be input together into a pre-trained intention recognition model, and a recognition result of whether the first instant message matches the target intention information may be obtained.
The first instant message is matched with the target intention information, specifically, the first instant message hits the intention of the target intention information.
Optionally, the obtained information matched with the target intention information may be a knowledge term, and correspondingly, the manner of obtaining the information matched with the target intention information may be that after the first instant information is determined to be matched with the target intention, the knowledge term matched with the target intention information is obtained in a preset knowledge graph.
It is understood that different intent information may be associated with different types of knowledge to obtain different types of knowledge vocabulary entries. As described above, if the target intention information is a paraphrasing demand, the encyclopedia map may be queried to obtain a matched paraphrasing encyclopedia entry; if the target intention information is the requirement of the person to be found, the contact person map can be queried, and matched contact way entries are obtained; if the target intention information is the opening requirement of the application program, the applet map can be queried to obtain matched applet entries and the like.
In a specific embodiment, assuming that the target entity is "CTR" and the target intention information is "paraphrase requirement for CTR", the information matched with the target intention information may be: CTR (Click-Through-Rate) is used to represent the proportion of the number of times a user clicks and enters a website to the total number of times the website is searched.
And 140, pushing information to the target terminal.
In this embodiment, the target terminal may be a terminal or a portable device (for example, a bracelet or a watch) used by the chat user who inputs the first instant message, or may be a terminal or a portable device used by all chat users in the instant messaging scenario (for example, a one-to-one chat scenario or a group chat scenario).
In this step, when pushing information to the target terminal, optionally, the information may be popped up to the target terminal directly in the chat application; or a browsing floating layer matched with the information can be generated, and the browsing floating layer is popped up to the target terminal.
According to the technical scheme, the input first instant message is subjected to intention recognition, the target entity is obtained, the target intention information corresponding to the target entity is obtained, the information matched with the target intention information is obtained in response to the fact that the first instant message is matched with the target intention information, and the information is pushed to the target terminal.
Fig. 2 is a flow chart of another information pushing method according to an embodiment of the disclosure, which is a further refinement of the foregoing technical solution, where the technical solution in this embodiment may be combined with one or more of the foregoing implementations. Specifically, referring to fig. 2, the method specifically includes the following steps:
step 210, performing entity identification on the input first instant message to obtain a target entity.
Step 220, determining a target entity type corresponding to the target entity according to the mapping relation between the entity and the entity type.
In this embodiment, the entity library includes a plurality of entities and entity types corresponding to the entities, and if the entity library includes an entity consistent with or overlapping with a target entity, the entity type corresponding to the entity may be determined according to a mapping relationship between each entity and the entity type stored in advance, and the entity type is taken as the target entity type corresponding to the target entity.
In a specific embodiment, it is assumed that the following entities are stored in an entity library: "Liu xx" and "Zhou newspaper", wherein the entity type corresponding to "Liu xx" is "personnel" and the entity type corresponding to "Zhou newspaper" is "APP". If the first instant message is "recall write week report", the target entity obtained after the identification by using the entity library may be "week report", and because the entity type corresponding to the entity "week report" is "APP", the target entity type corresponding to the first instant message "recall write week report" is also "APP".
Step 230, determining target intention information according to the target entity and the target entity type, wherein the target entity type is used for describing the intention of the target intention information.
In this embodiment, optionally, the target entity and the target entity type may be combined to obtain the target intention information.
In a specific embodiment, it is assumed that the target intention information is "weekly report, APP", where the target intention information is used to indicate an intention of opening an application program of "weekly report";
in another specific embodiment, it is assumed that the target intention information is "Liuxx, personnel", where the target intention information is used to indicate an intention or the like to acquire a person contact of "Liuxx".
The method has the advantages that the target intention information is obtained by combining the target entity and the target entity type, the intention corresponding to the target intention information can be accurately determined, and the accuracy of the information pushing result is improved.
Step 240, in response to determining that the first instant message matches the target intent message, obtaining information matching the target intent message.
In this embodiment, determining that the first instant message matches the target intention information may include: the target intention information and the first instant information are input into a pre-trained intention recognition model together; and determining whether the first instant message is matched with the target intention message according to the output result of the intention recognition model.
The intention recognition model can be obtained by training a neural network model through a plurality of training samples, and after target intention information and first instant information are input into the intention recognition model, the intention recognition model can calculate the semantic association degree between the first instant information and the target intention information. If the semantic association is higher, the first instant message can be determined to be matched with the target intention information, namely the first instant message hits the intention of the target intention information, and in this case, the result output by the intention recognition model can be a matched tag; otherwise, if the semantic association degree is low, it may be determined that the first instant message does not match the target intention information, that is, the first instant message does not hit the intention of the target intention information, in which case the result output by the intention recognition model may be a non-matching tag.
In a specific embodiment, assuming that the target intention information is "weekly report, APP", the first instant message is "recall and write weekly report", and since the first instant message includes the intention of opening the weekly report application program and writing the weekly report, after the target intention information and the first instant message are input together into the intention recognition model, the result output by the intention recognition model may be 1 (i.e. match the tag).
In another specific embodiment, assuming that the target intention information is "weekly report, APP", the first instant information is "weekly report, please follow in time", because the first instant information only refers to weekly report and does not include the intention of opening the weekly report application program, after the target intention information and the first instant information are input together into the intention recognition model, the result output by the intention recognition model may be 0 (i.e. a non-matching label).
The method has the advantages that whether the first instant message is matched with the target intention message or not can be quickly determined through the intention recognition model, and therefore the pushing efficiency of the message can be improved.
In this embodiment, the intent recognition model is obtained by fine tuning the knowledge-enhanced semantic representation ERNIE model using a training sample set obtained in advance. Wherein, each training sample comprises: intent information, input information and labeling results of whether the input information is matched with the intent information.
In this embodiment, the training sample set may include a plurality of training samples, and each training sample may include instant information (i.e., input information) input by each chat user collected in advance, intention information determined according to the input information, and a labeling result of whether the input information matches the intention information. Wherein corresponding intent information may be determined from the input information in the same manner as in steps 210-230.
In a specific embodiment, assuming that the input information in a certain training sample is "the question finds a Liu xx bar, he has clearer details," the corresponding intention information is "Liu xx, personnel," and since the input information contains intention about finding the personnel information of "Liu xx," the labeling result corresponding to the input information and the intention information can be set to 1.
Assuming that the input information in another training sample is "Liu xx is already online, I prepare to be online" and the corresponding intention text is "Liu xx, personnel", since the input information does not contain intention about finding the personnel information of "Liu xx", the labeling result corresponding to the input information and the intention information can be set to 0.
In this embodiment, the preset pre-training model may be trained by using the training sample set, and model parameters of the pre-training model may be adjusted according to the training result, so as to obtain the intent recognition model. The pre-training model may be an ERNIE model, among others. The method has the advantages that the ERNIE model has strong semantic modeling capability, and the training sample set matched with the instant messaging scene is used for training the ERNIE model, so that the intention recognition model can accurately recognize the intention of the chat user on different types of information in the instant messaging scene, and further the communication efficiency among different chat users is improved.
Step 250, pushing information to the target terminal.
According to the technical scheme, the target entity is obtained from the input first instant message, the target entity type corresponding to the target entity is determined according to the mapping relation between the entity and the entity type, the target intention information is determined according to the target entity and the target entity type, the information matched with the target intention information is obtained in response to the fact that the first instant message is matched with the target intention information, and the intention in the input instant message can be effectively and accurately identified by the aid of the operation means of pushing the information by the target terminal, so that the information matched with the intention can be accurately pushed, and the efficiency and accuracy of information pushing are improved.
Fig. 3a is a schematic flow chart of another method for pushing information according to an embodiment of the present disclosure, which is a further refinement of the foregoing technical solution, where the technical solution in the embodiment may be combined with one or more foregoing implementations. Specifically, referring to fig. 3a, the method specifically comprises the steps of:
step 310, performing at least one of the following processes on the input second instant message to obtain the first instant message: and deleting invalid instant messages and deleting content irrelevant characters in the instant messages.
In this embodiment, all information input by all chat users (i.e., the second instant information) may be obtained from the application program corresponding to the instant messaging scenario.
Since the second instant message generally contains information irrelevant to the intention recognition of the chat user, in order to improve the online prediction performance of the intention recognition model, some invalid instant messages need to be deleted in the second instant message, and characters irrelevant to the intention recognition (such as universal chat vocabulary, punctuation marks, etc.) in the instant messages, and the rest of the information is taken as the first instant message.
Optionally, assuming that the instant messaging scenario is an enterprise messaging scenario, the invalid instant message may be information unrelated to the work service, such as a link shared by entertainment activities, etc.; assuming that the communication scenario is a shopping communication scenario, the invalid instant message may be a message unrelated to shopping, such as a voting link, etc.
The advantage of this arrangement is that by removing invalid instant messages and characters from all instant messages, the workload of the intention recognition model can be reduced and the working efficiency of the intention recognition model can be improved.
In one implementation of this embodiment, the invalid instant message includes: instant messages in which the duty cycle of the target language character is less than or equal to a preset threshold value.
In one particular embodiment, chat users typically do not need to make intent recognition for large pieces of data or code-form information in instant messaging when communicating using an application. In this case, the duty ratio of the chinese characters to all the characters in the instant message may be counted, and whether the duty ratio is smaller than or equal to a preset threshold value may be determined, if so, the instant message may be used as an invalid instant message.
The method has the advantages that information irrelevant to the intention of the chat user can be effectively removed in the instant messaging scene, so that the online prediction performance of the intention recognition model is improved, and the communication efficiency between users in the instant messaging scene is improved.
In this embodiment, the deleting the content independent character in the instant message includes at least one of the following: deleting at least one of a uniform resource locator (Uniform Resource Locator, URL) symbol, a storage path and a storage catalog included in the instant message according to a preset regular matching rule; and when detecting that the instant message comprises the machine name, replacing the machine name by using a preset truncated character string.
In this embodiment, optionally, a preset regular expression may be used to match URL characters, storage paths and storage directories in the instant message, and the matched content may be deleted. In addition, the preset machine name can be used for comparing with the words in the instant message, and if the words which are the same as the preset machine name are included in the instant message, the words are replaced by the preset truncated character string.
The method has the advantages that the length of instant information can be reduced, the processing time of the intention recognition model on the content-independent characters is saved, and the working efficiency of the intention recognition model is improved.
Step 320, performing entity identification on the input first instant message to obtain a target entity.
Step 330, determining a target entity type corresponding to the target entity according to the mapping relationship between the entity and the entity type.
Step 340, determining the target intention information according to the target entity and the target entity type, wherein the target entity type is used for describing the intention of the target intention information.
Step 350, inputting the target intention information and the first instant information into a pre-trained intention recognition model, and acquiring information matched with the target intention information when the first instant information is determined to be matched with the target intention information according to the output result of the intention recognition model.
Step 360, pushing the information to the target terminal.
According to the technical scheme, invalid instant information in the second instant information and content irrelevant characters in the instant information are deleted, a target entity is obtained from the input first instant information, a target entity type corresponding to the target entity is determined according to a mapping relation between the entity and the entity type, target intention information is determined according to the target entity and the target entity type, the target intention information and the first instant information are input into a pre-trained intention recognition model together, when the first instant information is determined to be matched with the target intention information according to an output result of the intention recognition model, information matched with the target intention information is obtained, and the intention in the input instant information is effectively and accurately recognized by the technical means of pushing the information to the target terminal, so that the information matched with the intention can be pushed accurately, and the information pushing efficiency and accuracy are improved.
Based on the above embodiments, for a communication scenario involving sensitive information (e.g., an enterprise communication scenario), only limited input information is generally available in the related art, and due to the small number of training samples, insufficient learning of the intent recognition model is easily caused. In order to solve the above-mentioned problem, the present embodiment provides a training method of an intent recognition model, referring to fig. 3b, the training method includes:
step 301, performing sample expansion on a standard sample set of a first data scale by adopting a data enhancement algorithm to form the training sample set of a second data scale.
In this embodiment, optionally, the input information of the chat user may be obtained in the same manner as step 310, then the intention information corresponding to the input information is determined in the same manner as steps 320-340, and then the input information, the intention information, and the labeling result of whether the input information matches the intention information are used as one standard sample, and a standard sample set is obtained by processing a plurality of input information.
In this step, optionally, the input information in each standard sample may be fine-tuned (e.g. replacing or deleting random characters) to obtain a plurality of new samples, and the plurality of new samples are added to the standard sample set to obtain the training sample set.
The training sample set has the advantages that the scale of the training sample set can be enlarged, the problem that the training sample set is limited in number and the intention recognition model is insufficient to learn is avoided, the training effect of the intention recognition model can be improved, and the accuracy of an output result is ensured.
In one implementation of the present embodiment, the sample expansion is performed on the standard sample set by using a data enhancement algorithm, including at least one of the following: randomly masking characters in the standard sample to form a new sample; randomly removing at least one character from the standard sample to form a new sample; performing word vector hyponym replacement on at least one word in the standard sample to form a new sample; and inputting the standard sample containing the mask mark into the ERNIE model, and after the predicted character of the ERNIE model for the mask mark is obtained, replacing the mask mark in the standard sample by using the predicted character to form a new sample.
In a specific embodiment, a character may be randomly selected from the standard samples, and then deleted, or replaced with a predetermined mask, to obtain a new sample.
In another specific embodiment, the standard sample may be further segmented by using a preset word segmentation technique to obtain at least one original word corresponding to the standard sample, then a word vector of the original word is calculated, a target word closest to the original word (i.e., a hyponym corresponding to the original word) is obtained in a preset word segmentation library according to the word vector, and finally the original word is replaced by the target word to obtain a new sample. The word segmentation library stores a plurality of segmented words and word vectors corresponding to the segmented words respectively in advance.
In another specific embodiment, because the ERNIE model has a strong semantic modeling capability, after a standard sample containing the mask identifier is input into the ERNIE model, the ERNIE model predicts the mask identifier according to the semantics of the standard sample, and a predicted character corresponding to the mask identifier is obtained. Alternatively, the masking identifier in the standard sample may be replaced with the predicted character to obtain a new sample.
The data enhancement of each pair of standard sample sets is to convert input information in the standard sample by a pointer, and then generate a new sample by combining with intention information of the original input information and a labeling result.
The method has the advantages that the standard sample is processed by using at least one mode, so that the standard sample set can be expanded under the condition that the semantics of the standard sample are not influenced, the ERNIE model is trained by using a large-scale training sample set, the training effect of the intention recognition model can be improved, and the accuracy of an output result is ensured.
And 302, performing feature coding on a currently input target training sample through the ERNIE model to obtain a target sample feature set corresponding to the target training sample.
In this embodiment, the target training samples in the training sample set may be input into the ERNIE model. Wherein the ERNIE model comprises: the device comprises a coding module and a classifying module, wherein the classifying module comprises: a full connection layer and a logistic regression layer.
In this step, optionally, the current input target training sample may be feature-coded by the last layer (i.e., last feature extraction layer) of the coding module to obtain a target sample feature set corresponding to the target training sample, where the target sample feature set includes multiple target sample features.
In a specific embodiment, fig. 3c is a schematic structural diagram of an ERNIE model, and as shown in fig. 3c, the target training sample may include target intention information (including fields Tok1 and Tok2 and … … TokN), target input information (including fields Sok1 and Sok and … … SokN), and a labeling result of whether the target input information matches the target intention information. After the target training sample is input into the ERNIE model, a coding module in the ERNIE model calculates the characteristics respectively corresponding to the target intention information and the target input information in the target training sample. Taking fig. 3c as an example, the features corresponding to the target intention information include t_1 and t_ … … t_n, and the features corresponding to the target input information include s_1 and s_ … … s_n. Wherein each field in the target intention information and the target input information corresponds to a feature.
In this step, optionally, the features corresponding to the target intention information and the target input information may be used together as the target sample features corresponding to the target training sample.
In one implementation manner of the embodiment, feature encoding is performed on a currently input target training sample through an ERNIE model to obtain a target sample feature set corresponding to the target training sample, including: and carrying out feature coding on target input information in a target training sample through an ERNIE model to obtain a target sample feature set corresponding to the target training sample.
In a specific embodiment, taking fig. 3c as an example, after the feature calculation is completed, the last feature extraction layer of the encoding module obtains the feature corresponding to the target input information (i.e. s_1, s_ … … s_n) as the target sample feature corresponding to the target training sample.
In this embodiment, since the content of the intention information is generally fixed, the characteristic is relatively single, and in the training process of the ERNIE model, the characteristic of the input information really has a hard effect on the accuracy of the output result. Therefore, in order to improve the training efficiency of the ERNIE model, the features of the intention information may be removed, and only the features of the input information are acquired as sample features. The method has the advantages that the calculation amount of the ERNIE model can be reduced, and the training efficiency of the ERNIE model is improved.
In a particular embodiment, the intent information may be extended to a fixed length (e.g., 16 bits) to facilitate the extraction of features of the input information by the ERNIE model.
And 303, performing twice random omission processing on each target sample feature included in the target sample feature set through the ERNIE model to obtain a first feature set and a second feature set.
In this embodiment, each target sample feature may be input to a classification module in the ERNIE model, and the full-connection layer in the classification module may perform a first random missing process (dropout) on each target sample feature, that is, ignore a part of features in all target sample features (make a feature value of a part of features be 0), and use the remaining other part of features as the first feature set.
And then inputting each target sample characteristic into a classification module in the ERNIE model again, wherein a full-connection layer in the classification module can perform a second random omission process (dropout) on each target sample characteristic, namely ignoring part of the characteristics in all target sample characteristics, and taking the rest of the other characteristics as a second characteristic set.
And 304, processing the first feature set and the second feature set by adopting a preset classification algorithm through the ERNIE model to obtain a first distribution function and a second distribution function.
In this step, the first feature set may be input to the logistic regression layer, and a preset classification algorithm is adopted by the logistic regression layer to calculate a distribution function corresponding to the first feature set, so as to obtain a first distribution function.
And inputting the second feature set into the logistic regression layer again, and calculating a distribution function corresponding to the second feature set by adopting a preset classification algorithm through the logistic regression layer to obtain a second distribution function.
And 305, calculating the relative entropy corresponding to the target training sample according to the first distribution function and the second distribution function through the ERNIE model.
In this step, optionally, a distance between the first distribution function and the second distribution function may be calculated, and the distance is taken as a relative entropy corresponding to the target training sample.
And 306, adjusting model parameters according to the relative entropy.
In this embodiment, if the value of the relative entropy is large, the training result of the ERNIE model may be considered to be poor, and the parameters of the ERNIE model may be adjusted to reduce the value of the relative entropy.
The method has the advantages that through carrying out random omission processing on the target sample characteristics, interaction among the characteristics can be reduced, dependence of model training results on local characteristics is reduced, and therefore generalization capability of an intention recognition model can be improved; secondly, the target sample characteristics are continuously input into the classification module twice, and parameter adjustment is carried out on the ERNIE model according to the relative entropy corresponding to the target training sample, so that the training effect of the intention recognition model can be improved, and the accuracy of the output result of the intention recognition model is ensured.
The embodiment of the disclosure also provides an information pushing device, which is used for executing the information pushing method.
Fig. 4 is a block diagram of an information pushing device 400 according to an embodiment of the present disclosure, where the device includes: an entity acquisition module 410, an intent acquisition module 420, an information acquisition module 430, and an information push module 440.
The entity obtaining module 410 is configured to perform entity identification on the input first instant message, and obtain a target entity;
an intention obtaining module 420, configured to obtain target intention information corresponding to the target entity;
an information obtaining module 430, configured to obtain information matched with the target intention information in response to determining that the first instant information matches the target intention information;
and the information pushing module 440 is configured to push the information to the target terminal.
According to the technical scheme, the input first instant message is subjected to entity identification, the target entity is obtained, the target intention information corresponding to the target entity is obtained, the information matched with the target intention information is obtained in response to the fact that the first instant message is matched with the target intention information, and the information is pushed to the target terminal.
On the basis of the above embodiments, the intention obtaining module 420 includes:
the entity type determining unit is used for determining a target entity type corresponding to the target entity according to the mapping relation between the entity and the entity type;
and the intention information determining unit is used for determining the target intention information according to the target entity and the target entity type, wherein the target entity type is used for describing the intention of the target intention information.
The information acquisition module 430 includes:
the information input unit is used for inputting the target intention information and the first instant information together into a pre-trained intention recognition model;
the matching result determining unit is used for determining whether the first instant message is matched with the target intention message according to the output result of the intention recognition model;
the intention recognition model is obtained by fine tuning a semantic representation ERNIE model with enhanced knowledge by using a training sample set acquired in advance;
each training sample comprises: intent information, input information and labeling results of whether the input information is matched with the intent information.
The apparatus further comprises: the training sample set acquisition unit is used for carrying out sample expansion on the standard sample set with the first data scale by adopting a data enhancement algorithm to form the training sample set with the second data scale;
The training sample set acquisition unit includes:
the masking processing subunit is used for carrying out random masking processing on characters in the standard sample to form a new sample; a character removing subunit, configured to randomly remove at least one character from the standard sample, to form a new sample; the word segmentation replacing subunit is used for carrying out word vector hyponym replacement on at least one word in the standard sample to form a new sample; the standard sample input subunit is used for inputting a standard sample containing a mask identifier into the ERNIE model, and after a predicted character of the ERNIE model on the mask identifier is obtained, replacing the mask identifier in the standard sample by using the predicted character to form a new sample.
The information input unit includes:
the feature coding subunit is used for carrying out feature coding on a currently input target training sample through the ERNIE model to obtain a target sample feature set corresponding to the target training sample; the omission processing subunit is used for carrying out twice random omission processing on each target sample feature included in the target sample feature set through the ERNIE model to obtain a first feature set and a second feature set; the distribution function determining subunit is used for processing the first feature set and the second feature set through the ERNIE model by adopting a preset classification algorithm to obtain a first distribution function and a second distribution function; the relative entropy calculating subunit is used for calculating relative entropy corresponding to the target training sample according to the first distribution function and the second distribution function through the ERNIE model; the model parameter adjustment subunit is used for adjusting the model parameters according to the relative entropy;
The feature encoding subunit includes:
and the input information coding subunit is used for carrying out feature coding on the target input information in the target training sample through the ERNIE model to obtain a target sample feature set corresponding to the target training sample.
The information pushing device further comprises: the information processing module is used for processing the input second instant message at least one of the following to obtain the first instant message: deleting invalid instant messages; deleting content independent characters in the instant message;
the invalid instant message includes: instant messages in which the duty cycle of the target language character is less than or equal to a preset threshold value.
The information processing module includes: the information deleting subunit is used for deleting at least one of a uniform resource locator, a storage path and a storage catalog which are included in the instant message according to a preset regular matching rule; and the character string replacing subunit is used for replacing the machine name by using a preset truncated character string when the fact that the machine name is included in the instant message is detected.
The information pushing device provided by the embodiment of the disclosure can execute the information pushing method provided by any embodiment of the disclosure, and has the corresponding functional modules and beneficial effects of the execution method.
According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.
Fig. 5 illustrates a schematic block diagram of an example electronic device 500 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.
As shown in fig. 5, the apparatus 500 includes a computing unit 501 that can perform various suitable actions and processes according to a computer program stored in a Read Only Memory (ROM) 502 or a computer program loaded from a storage unit 505 into a Random Access Memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, ROM 502, and RAM 503 are connected to each other by a bus 504. An input/output (I/O) interface 505 is also connected to bus 504.
Various components in the device 500 are connected to the I/O interface 505, including: an input unit 506 such as a keyboard, a mouse, etc.; an output unit 507 such as various types of displays, speakers, and the like; a storage unit 505 such as a magnetic disk, an optical disk, or the like; and a communication unit 509 such as a network card, modem, wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.
The computing unit 501 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 501 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing AI chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the respective methods and processes described above, such as a push method of information. For example, in some embodiments, the method of pushing information may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as the storage unit 505. In some embodiments, part or all of the computer program may be loaded and/or installed onto the device 500 via the ROM 502 and/or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the push method of information described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform the push method of information in any other suitable way (e.g. by means of firmware).
Various implementations of the systems and techniques described here above can be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), AI system-on-chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.
Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), blockchain networks, and the internet.
The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service are overcome.
It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the technical solutions provided by the present disclosure are achieved, and are not limited herein.
The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.
Claims (22)
1. A method of pushing information, the method comprising:
entity identification is carried out on the input first instant message, and a target entity is obtained; the target entity is information which is used or understood by the chat user and is other than a preset universal chat vocabulary in the first instant message;
acquiring target intention information corresponding to the target entity; the target entity corresponds to a plurality of target intention information;
In response to determining that the first instant message matches the target intent message, obtaining information matching the target intent message; pushing the information to a target terminal;
the first instant message is chat information input by each chat user respectively acquired in a chat application program in an instant communication session scene;
determining that the first instant message matches the target intent information includes:
the target intention information and the first instant message are input into a pre-trained intention recognition model together; determining whether the first instant message is matched with the target intention message according to an output result of the intention recognition model; the intention recognition model is used for calculating semantic association degree between the first instant message and the target intention message.
2. The method of claim 1, wherein the obtaining target intent information corresponding to the target entity comprises:
determining a target entity type corresponding to the target entity according to the mapping relation between the entity and the entity type;
and determining the target intention information according to the target entity and the target entity type, wherein the target entity type is used for describing the intention of the target intention information.
3. The method of claim 1, wherein,
the intention recognition model is obtained by fine tuning a semantic representation ERNIE model with knowledge enhancement by using a training sample set obtained in advance;
each training sample comprises: intent information, input information and labeling results of whether the input information is matched with the intent information.
4. A method according to claim 3, wherein obtaining a training sample set comprises:
and performing sample expansion on the standard sample set with the first data size by adopting a data enhancement algorithm to form the training sample set with the second data size.
5. The method of claim 4, wherein sample expansion of the standard sample set using a data enhancement algorithm comprises at least one of:
randomly masking characters in the standard sample to form a new sample;
randomly removing at least one character from the standard sample to form a new sample;
performing word vector hyponym replacement on at least one word in the standard sample to form a new sample; and
inputting a standard sample containing a mask identifier into the ERNIE model, and after a predicted character of the ERNIE model for the mask identifier is obtained, replacing the mask identifier in the standard sample by using the predicted character to form a new sample.
6. The method of claim 4, wherein refining the ERNIE model using the pre-acquired training sample set to obtain the intent recognition model comprises:
performing feature coding on a currently input target training sample through the ERNIE model to obtain a target sample feature set corresponding to the target training sample;
performing twice random omission processing on each target sample feature included in the target sample feature set through the ERNIE model to obtain a first feature set and a second feature set;
processing the first feature set and the second feature set by adopting a preset classification algorithm through the ERNIE model to obtain a first distribution function and a second distribution function;
calculating relative entropy corresponding to the target training sample according to the first distribution function and the second distribution function through the ERNIE model;
and adjusting model parameters according to the relative entropy.
7. The method of claim 6, wherein feature encoding, by the ERNIE model, a currently input target training sample to obtain a target sample feature set corresponding to the target training sample, comprises:
And carrying out feature coding on target input information in the target training sample through the ERNIE model to obtain a target sample feature set corresponding to the target training sample.
8. The method of any of claims 1-7, wherein prior to the obtaining the target entity from the entered first instant message, further comprising:
processing the input second instant message to obtain the first instant message, wherein the processing comprises at least one of the following steps:
deleting invalid instant messages;
deleting the content independent character in the instant message.
9. The method of claim 8, wherein the invalid instant message comprises: instant messages in which the duty cycle of the target language character is less than or equal to a preset threshold value.
10. The method of claim 8, wherein the deleting content independent characters in the instant message comprises at least one of:
deleting at least one of a uniform resource locator, a storage path and a storage catalog included in the instant message according to a preset regular matching rule;
and when detecting that the instant message comprises the machine name, replacing the machine name by using a preset truncated character string.
11. An information pushing apparatus, the apparatus comprising:
The entity acquisition module is used for carrying out entity identification on the input first instant message to acquire a target entity; the target entity is information which is used or understood by the chat user and is other than a preset universal chat vocabulary in the first instant message;
the intention acquisition module is used for acquiring target intention information corresponding to the target entity; the target entity corresponds to a plurality of target intention information;
the information acquisition module is used for responding to the fact that the first instant information is matched with the target intention information, and acquiring information matched with the target intention information;
the information pushing module is used for pushing the information to the target terminal;
the first instant message is chat information input by each chat user respectively acquired in a chat application program in an instant communication session scene;
determining that the first instant message matches the target intent information includes: the target intention information and the first instant message are input into a pre-trained intention recognition model together; determining whether the first instant message is matched with the target intention message according to an output result of the intention recognition model; the intention recognition model is used for calculating semantic association degree between the first instant message and the target intention message.
12. The apparatus of claim 11, wherein the intent acquisition module includes:
the entity type determining unit is used for determining a target entity type corresponding to the target entity according to the mapping relation between the entity and the entity type;
and the intention information determining unit is used for determining the target intention information according to the target entity and the target entity type, wherein the target entity type is used for describing the intention of the target intention information.
13. The apparatus of claim 11, wherein,
the intention recognition model is obtained by fine tuning a semantic representation ERNIE model with knowledge enhancement by using a training sample set obtained in advance;
each training sample comprises: intent information, input information and labeling results of whether the input information is matched with the intent information.
14. The apparatus of claim 13, the apparatus further comprising:
the training sample set acquisition unit is used for carrying out sample expansion on the standard sample set with the first data scale by adopting a data enhancement algorithm to form the training sample set with the second data scale.
15. The apparatus of claim 14, wherein the training sample set acquisition unit comprises:
The masking processing subunit is used for carrying out random masking processing on characters in the standard sample to form a new sample;
a character removing subunit, configured to randomly remove at least one character from the standard sample, to form a new sample;
the word segmentation replacing subunit is used for carrying out word vector hyponym replacement on at least one word in the standard sample to form a new sample;
the standard sample input subunit is used for inputting a standard sample containing a mask identifier into the ERNIE model, and after a predicted character of the ERNIE model on the mask identifier is obtained, replacing the mask identifier in the standard sample by using the predicted character to form a new sample.
16. The apparatus of claim 14, wherein the information input unit comprises:
the feature coding subunit is used for carrying out feature coding on a currently input target training sample through the ERNIE model to obtain a target sample feature set corresponding to the target training sample;
the omission processing subunit is used for carrying out twice random omission processing on each target sample feature included in the target sample feature set through the ERNIE model to obtain a first feature set and a second feature set;
The distribution function determining subunit is used for processing the first feature set and the second feature set through the ERNIE model by adopting a preset classification algorithm to obtain a first distribution function and a second distribution function;
the relative entropy calculating subunit is used for calculating relative entropy corresponding to the target training sample according to the first distribution function and the second distribution function through the ERNIE model;
and the model parameter adjustment subunit is used for adjusting the model parameters according to the relative entropy.
17. The apparatus of claim 16, wherein the feature encoding subunit comprises:
and the input information coding subunit is used for carrying out feature coding on the target input information in the target training sample through the ERNIE model to obtain a target sample feature set corresponding to the target training sample.
18. The apparatus according to any one of claims 11-17, further comprising:
the information processing module is used for processing the input second instant message at least one of the following to obtain the first instant message:
deleting invalid instant messages;
deleting the content independent character in the instant message.
19. The apparatus of claim 18, the invalid instant message comprising: instant messages in which the duty cycle of the target language character is less than or equal to a preset threshold value.
20. The apparatus of claim 18, the information processing module comprising:
the information deleting subunit is used for deleting at least one of a uniform resource locator, a storage path and a storage catalog which are included in the instant message according to a preset regular matching rule;
and the character string replacing subunit is used for replacing the machine name by using a preset truncated character string when the fact that the machine name is included in the instant message is detected.
21. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-10.
22. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111544848.7A CN114244795B (en) | 2021-12-16 | 2021-12-16 | Information pushing method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111544848.7A CN114244795B (en) | 2021-12-16 | 2021-12-16 | Information pushing method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114244795A CN114244795A (en) | 2022-03-25 |
| CN114244795B true CN114244795B (en) | 2024-02-09 |
Family
ID=80757169
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111544848.7A Active CN114244795B (en) | 2021-12-16 | 2021-12-16 | Information pushing method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114244795B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115086269A (en) * | 2022-06-15 | 2022-09-20 | 中银金融科技有限公司 | Address book query method and device based on enterprise WeChat |
| CN115358223A (en) * | 2022-09-05 | 2022-11-18 | 北京百度网讯科技有限公司 | Information prediction method, device, equipment and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108153901A (en) * | 2018-01-16 | 2018-06-12 | 北京百度网讯科技有限公司 | The information-pushing method and device of knowledge based collection of illustrative plates |
| WO2019011356A1 (en) * | 2017-07-14 | 2019-01-17 | Cognigy Gmbh | Method for conducting dialog between human and computer |
| CN110888968A (en) * | 2019-10-15 | 2020-03-17 | 浙江省北大信息技术高等研究院 | Customer service dialogue intention classification method and device, electronic equipment and medium |
| CN110909137A (en) * | 2019-10-12 | 2020-03-24 | 平安科技（深圳）有限公司 | Information pushing method and device based on man-machine interaction and computer equipment |
| CN111104495A (en) * | 2019-11-19 | 2020-05-05 | 深圳追一科技有限公司 | Information interaction method, device, equipment and storage medium based on intention recognition |
| CN111161740A (en) * | 2019-12-31 | 2020-05-15 | 中国建设银行股份有限公司 | Intention recognition model training method, intention recognition method and related device |
| CN111666415A (en) * | 2020-06-28 | 2020-09-15 | 深圳壹账通智能科技有限公司 | Topic clustering method and device, electronic equipment and storage medium |
| CN111753056A (en) * | 2020-06-28 | 2020-10-09 | 北京百度网讯科技有限公司 | Information pushing method and device, computing equipment and computer readable storage medium |
| CN113391874A (en) * | 2020-03-12 | 2021-09-14 | 腾讯科技（深圳）有限公司 | Virtual machine detection countermeasure method and device, electronic equipment and storage medium |
-
2021
- 2021-12-16 CN CN202111544848.7A patent/CN114244795B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019011356A1 (en) * | 2017-07-14 | 2019-01-17 | Cognigy Gmbh | Method for conducting dialog between human and computer |
| CN108153901A (en) * | 2018-01-16 | 2018-06-12 | 北京百度网讯科技有限公司 | The information-pushing method and device of knowledge based collection of illustrative plates |
| CN110909137A (en) * | 2019-10-12 | 2020-03-24 | 平安科技（深圳）有限公司 | Information pushing method and device based on man-machine interaction and computer equipment |
| WO2021068321A1 (en) * | 2019-10-12 | 2021-04-15 | 平安科技（深圳）有限公司 | Information pushing method and apparatus based on human-computer interaction, and computer device |
| CN110888968A (en) * | 2019-10-15 | 2020-03-17 | 浙江省北大信息技术高等研究院 | Customer service dialogue intention classification method and device, electronic equipment and medium |
| CN111104495A (en) * | 2019-11-19 | 2020-05-05 | 深圳追一科技有限公司 | Information interaction method, device, equipment and storage medium based on intention recognition |
| CN111161740A (en) * | 2019-12-31 | 2020-05-15 | 中国建设银行股份有限公司 | Intention recognition model training method, intention recognition method and related device |
| CN113391874A (en) * | 2020-03-12 | 2021-09-14 | 腾讯科技（深圳）有限公司 | Virtual machine detection countermeasure method and device, electronic equipment and storage medium |
| CN111666415A (en) * | 2020-06-28 | 2020-09-15 | 深圳壹账通智能科技有限公司 | Topic clustering method and device, electronic equipment and storage medium |
| CN111753056A (en) * | 2020-06-28 | 2020-10-09 | 北京百度网讯科技有限公司 | Information pushing method and device, computing equipment and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114244795A (en) | 2022-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111625635A (en) | Question-answer processing method, language model training method, device, equipment and storage medium | |
| CN114244795B (en) | Information pushing method, device, equipment and medium | |
| CN114363019B (en) | Training method, device, equipment and storage medium for phishing website detection model | |
| CN113836925A (en) | Training method and device for pre-training language model, electronic equipment and storage medium | |
| CN112528641A (en) | Method and device for establishing information extraction model, electronic equipment and readable storage medium | |
| CN112560461A (en) | News clue generation method and device, electronic equipment and storage medium | |
| CN112528146B (en) | Content resource recommendation method and device, electronic equipment and storage medium | |
| CN113904943A (en) | Account detection method and device, electronic equipment and storage medium | |
| CN114254650A (en) | Information processing method, device, equipment and medium | |
| CN116383382A (en) | Sensitive information identification method and device, electronic equipment and storage medium | |
| CN115600592A (en) | Method, device, equipment and medium for extracting key information of text content | |
| CN115101069A (en) | Voice control method, device, equipment, storage medium and program product | |
| CN115098729A (en) | Video processing method, sample generation method, model training method and device | |
| CN114417862A (en) | Text matching method, and training method and device of text matching model | |
| CN115248890A (en) | User interest portrait generation method and device, electronic equipment and storage medium | |
| CN112784600A (en) | Information sorting method and device, electronic equipment and storage medium | |
| CN115131709B (en) | Video category prediction method, training method and device for video category prediction model | |
| CN116069914B (en) | Training data generation method, model training method and device | |
| CN116244740B (en) | Log desensitization method and device, electronic equipment and storage medium | |
| CN113377922B (en) | Method, device, electronic equipment and medium for matching information | |
| CN113377921B (en) | Method, device, electronic equipment and medium for matching information | |
| CN114329230B (en) | Information generation method and device | |
| CN113238765B (en) | Method, device, equipment and storage medium for distributing small program | |
| CN116628167B (en) | Response determination method and device, electronic equipment and storage medium | |
| CN112988688A (en) | Picture sharing method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |