Information timeliness identification, model training, push method, device and medium

By performing statement segmentation and feature data processing on the information, and using the time-limit identification model to accurately identify the time-limiting of information push in the existing technology, the problem of inaccurate time-limiting in information push in the existing technology is solved, and the quality of information push and user experience are improved.

CN114330295BActive Publication Date: 2024-11-26TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110892912.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-08-04
Publication Date
2024-11-26
Estimated Expiration
2041-08-04

AI Technical Summary

Technical Problem

The lack of effective information timeliness identification methods in the prior art leads to inaccurate information push timeliness, resulting in useful information being removed prematurely or useless information being pushed, affecting the user experience.

Method used

By obtaining the target text data from the information to be identified, performing statement segmentation processing, determining the statement feature data, and inputting it into the time-limiting recognition model, obtaining the time-limiting recognition results of the statement, comprehensively judging the time-limiting length of the information, and accurately determining the time-limiting of the information.

Benefits of technology

Improve the quality and user experience of information push, ensure that information is pushed within the appropriate time, and reduce the occurrence of misleading and pushing expired information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114330295B_ABST
    Figure CN114330295B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device and medium for identifying the timeliness of information, model training and pushing, wherein the method for identifying the timeliness of information obtains target text data from the information to be identified, performs sentence segmentation processing on the target text data to obtain multiple sentences, and determines the sentence feature data of each sentence; then, based on the timeliness identification model, the sentence timeliness identification result is determined according to the sentence feature data of each sentence, and the shorter timeliness corresponding to the sentence is determined as the timeliness identification result of the information. The timeliness identification method divides the information into multiple sentences, and comprehensively judges the timeliness of the information based on the sentence-level timeliness identification results, which can determine the timeliness of the information more accurately, which is conducive to improving the quality of information push and optimizing user experience. The present application can be widely used in the field of information technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of information technology, and in particular to a method, device and medium for identifying the timeliness of information, training a model and pushing it. Background Art

[0002] In recent years, with the rapid development of network media technology, the scale of various types of information has been increasing day by day, and Internet platforms can push this information to users for browsing. Among them, some information has time limits and needs to be pushed to users within a specified time, otherwise it is easy to lose browsing value, and may even have a misleading negative effect, affecting user experience.

[0003] In the related art, the push timeliness is generally determined according to the category of information. After receiving the information, the platform determines the timeliness according to the category, and then takes the time of the first push as the starting point, and removes the information after a preset period of time. However, in reality, the timeliness of different information varies greatly. The push timeliness determined by the above method often causes useful information to be removed prematurely or useless information to be pushed. In summary, the related art lacks a more practical information timeliness identification method, resulting in less than ideal information push results. Summary of the invention

[0004] The purpose of this application is to solve one of the technical problems existing in the prior art to at least a certain extent.

[0005] To this end, one purpose of an embodiment of the present application is to provide a method for identifying the timeliness of information, which can determine the timeliness of information more accurately and is conducive to improving the quality of information push.

[0006] Another purpose of an embodiment of the present application is to provide a method for training an information timeliness recognition model. The timeliness recognition model obtained by the training method can effectively and accurately identify the available timeliness of different information.

[0007] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present application include:

[0008] In one aspect, an embodiment of the present application provides a method for identifying the timeliness of information, the method comprising the following steps:

[0009] Acquire target text data from the information to be recognized;

[0010] Performing sentence segmentation processing on the target text data to obtain a plurality of sentences, and determining first sentence feature data corresponding to each of the sentences;

[0011] Inputting the first sentence feature data corresponding to each of the sentences into a time-limited recognition model to obtain a first time-limited recognition result corresponding to each of the sentences; the first time-limited recognition result is used to characterize the effective time of the sentence;

[0012] A second timeliness identification result corresponding to the information is determined according to the first timeliness identification result representing a shorter effective time; the second timeliness identification result is used to represent the effective time of the information.

[0013] On the other hand, an embodiment of the present application provides a method for training a time-sensitive information recognition model, the method comprising the following steps:

[0014] Acquire a batch of sample sentences and time-limit tags corresponding to the sample sentences; the time-limit tags are used to indicate the effective time of the sample sentences;

[0015] Determining second sentence feature data of each of the sample sentences;

[0016] Inputting the second sentence feature data into a time-limit recognition model to obtain a fourth time-limit recognition result of the sample sentence; the fourth time-limit recognition result is used to characterize the effective time of the sample sentence;

[0017] Determining a loss value of training according to the fourth timeliness recognition result and the timeliness label;

[0018] The parameters of the aging recognition model are updated according to the loss value to obtain a trained aging recognition model.

[0019] On the other hand, an embodiment of the present application provides a method for pushing information, the method comprising the following steps:

[0020] Get the information to be pushed;

[0021] Performing timeliness identification on the information by the aforementioned method to obtain a timeliness identification result;

[0022] Push the information according to the timeliness identification result; or push the information and the timeliness identification result.

[0023] On the other hand, an embodiment of the present application provides a device for identifying timeliness of information, including:

[0024] An acquisition module, used for acquiring target text data from the information to be identified;

[0025] A segmentation module, used for performing sentence segmentation processing on the target text data to obtain a plurality of sentences, and determining first sentence feature data corresponding to each of the sentences;

[0026] A recognition module, used for inputting the first sentence feature data corresponding to each of the sentences into a time-limit recognition model to obtain a first time-limit recognition result corresponding to each of the sentences; the first time-limit recognition result is used to characterize the effective time of the sentence;

[0027] The processing module is used to determine a second timeliness identification result corresponding to the information according to the first timeliness identification result which represents a shorter effective time; the second timeliness identification result is used to represent the effective time of the information.

[0028] On the other hand, an embodiment of the present application provides a computer device, including:

[0029] at least one processor;

[0030] at least one memory for storing at least one program;

[0031] When the at least one program is executed by the at least one processor, the at least one processor implements the aforementioned method for identifying the timeliness of information, the method for training a timeliness identification model for information, or the method for pushing information.

[0032] On the other hand, an embodiment of the present application also provides a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to implement the information timeliness identification method, information timeliness identification model training method or information push method mentioned above.

[0033] On the other hand, an embodiment of the present application also provides a computer program product or a computer program, which includes computer instructions, and the computer instructions are stored in the computer-readable storage medium mentioned above; the processor of the computer device mentioned above can read the computer instructions from the computer-readable storage medium mentioned above, and the processor executes the computer instructions, so that the computer device executes the information timeliness identification method, information timeliness identification model training method or information push method mentioned above.

[0034] The advantages and beneficial effects of the present invention will be partly given in the following description, partly become apparent from the following description, or be understood through the practice of the present invention:

[0035] The timeliness identification method for information provided in the embodiments of the present application obtains target text data from the information to be identified, performs sentence segmentation processing on the target text data to obtain multiple sentences, and determines the sentence feature data of each sentence; then, based on the timeliness identification model, the timeliness identification result of the sentence is determined according to the sentence feature data of each sentence, and the shorter timeliness corresponding to the sentence is determined as the timeliness identification result of the information. This timeliness identification method divides the information to be identified into multiple sentences, and comprehensively judges the timeliness of the information based on the sentence-level timeliness identification results, which can determine the timeliness of the information more accurately, which is conducive to improving the quality of information push and optimizing user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following introduction is made to the drawings of the embodiments of the present application or the related technical solutions in the prior art. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 A schematic diagram of a method for identifying the timeliness of information in related technologies;

[0038] Figure 2 A schematic diagram of an implementation environment of a method for identifying the timeliness of information provided in an embodiment of the present application;

[0039] Figure 3 A schematic diagram of a flow chart of a method for identifying the timeliness of information provided in an embodiment of the present application;

[0040] Figure 4 A schematic diagram of obtaining initial text data in a method for identifying the timeliness of information provided in an embodiment of the present application;

[0041] Figure 5 A schematic diagram of obtaining target text data in a method for identifying the timeliness of information provided in an embodiment of the present application;

[0042] Figure 6 A schematic diagram of the word segmentation principle in a method for identifying the timeliness of information provided in an embodiment of the present application;

[0043] Figure 7 A schematic diagram of determining a second timeliness identification result in a timeliness identification method for information provided in an embodiment of the present application;

[0044] Figure 8 A schematic diagram of a specific implementation process of a method for identifying the timeliness of information provided in an embodiment of the present application;

[0045] Fig. 9 A flowchart of a method for training a time-sensitive information recognition model provided in an embodiment of the present application;

[0046] Fig.10 A flowchart of an information recommendation method provided in an embodiment of the present application;

[0047] Fig.11 A schematic diagram of a terminal interface for applying an information recommendation method provided in an embodiment of the present application;

[0048] Fig.12 A schematic diagram of the structure of a device for identifying the timeliness of information provided in an embodiment of the present application;

[0049] Fig.13 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and are not to be construed as limitations on the present application. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0052] Before describing the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application are first described. The nouns and terms involved in the embodiments of the present application are subject to the following explanations.

[0053] Automatic Speech Recognition (ASR): This technology enables computers to "dictate" continuous speech spoken by different people, commonly known as "voice dictation machine". It is a technology that realizes the conversion of "sound" to "text" and can convert human voice information into text information.

[0054] Optical Character Recognition (OCR): This technology refers to the technology in which an electronic device (such as a scanner or digital camera) examines the characters in an image, determines their shape by detecting dark and light patterns, and then uses character recognition methods to translate the shape into text.

[0055] Timeliness: refers to the validity period of information. When the information is published beyond the timeliness, the accuracy of its content decreases, and its browsing value decreases accordingly, which may bring misleading negative effects to users.

[0056] The timeliness identification method of information, information recommendation method and timeliness identification model training method of information provided in the embodiments of the present application mainly involve artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level technology and software-level technology. Among them, basic artificial intelligence technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies; artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, machine learning / deep learning, autonomous driving and smart transportation.

[0057] The timeliness identification method of information, information recommendation method and timeliness identification model training method of information provided in the embodiment of the present application specifically relate to natural language processing technology and machine learning in the field of artificial intelligence. Among them, natural language processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science and mathematics. The natural language involved in this field is the language used by people in daily life, so it is also closely related to the study of linguistics. Natural language processing technology generally includes technologies such as text processing, semantic understanding, machine translation, robot question and answer, knowledge graph, etc. Machine learning (ML) is a multi-field interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or realize human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning (deep learning) usually includes artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0058] Specifically, the method provided in the embodiments of the present application can use natural language processing technology to process the target text data in the information, and combine the machine learning model to predict the timeliness identification results of the information. These methods can be executed in various application scenarios in the field of artificial intelligence: for example, if a user wants to know the timeliness of the information he receives, he can use an artificial intelligence system that can execute the timeliness identification method of the information in this application to identify the information; if the push platform wants to arrange the push priority or removal time of various types of information in an orderly manner, the artificial intelligence system can formulate a push strategy based on the information push method in the embodiments of the present application, so as to efficiently push information to users. It can be understood by those skilled in the art that the above application scenarios only serve as an example and do not mean to limit the actual application of the methods in the embodiments of the present application. In different application scenarios, the artificial intelligence system can use the methods provided in the embodiments of the present application to perform specified tasks.

[0059] Nowadays, the development of the Internet has profoundly changed people's lifestyles. A flood of information is constantly pouring into people's lives, entertainment and work. Among them, online media can push various types of information to users for browsing, broaden users' horizons and enrich users' cognition. However, at the same time, some information has time limits, so it needs to be pushed to users within a specified time. If the timeliness of information recommendations is inaccurate, it will cause browsing fatigue to users to a certain extent, and it is easy to mislead and may cause users to be disgusted.

[0060] The information push method in the related art is generally based on the category to which the information belongs and the rules set by humans. Figure 1 For the information 110 to be pushed, a simple judgment can be made on the category to which the information 110 to be pushed belongs through the information classification standard 120. For example, the category of the information can be judged by the source channel of the information, the features in the text or the title. After obtaining the category of the information 110 to be pushed, the information to be pushed is pushed according to the preset time limit rules. For example, when the information 110 to be pushed is sports-related information, the default push time is 3 days; when the information 110 to be pushed is movie-related information, the default push time is 7 days. Similarly, the default push time for social or financial information is 2 days, etc. These information will start counting from the first time they are pushed, and will be removed from the shelves and no longer pushed after the default push time is reached. However, in reality, the time limits of different information vary greatly. Figure 1 The push method shown simply estimates the push time according to the category of information, which often leads to useful information being removed prematurely and the utilization rate of some valuable information being low; or useless information is still pushed, which not only increases the push cost, but also the expired information that has lost its browsing value will affect the user experience. In summary, the related technology lacks a more practical information timeliness identification method, resulting in less than ideal information push results.

[0061] In view of this, an embodiment of the present application provides a method for identifying the timeliness of information, which obtains target text data from the information to be identified, performs sentence segmentation processing on the target text data to obtain multiple sentences, and determines the sentence feature data of each sentence; then, based on the timeliness identification model, the sentence timeliness identification result is determined according to the sentence feature data of each sentence, and the shorter timeliness corresponding to the sentence is determined as the timeliness identification result of the information. The timeliness identification method divides the information into multiple sentences, and comprehensively judges the timeliness of the information based on the sentence-level timeliness identification results, which can determine the timeliness of the information more accurately, which is conducive to improving the quality of information push and optimizing user experience.

[0062] The specific implementation of the embodiment of the present application is described in detail below in conjunction with the accompanying drawings.

[0063] Reference Figure 2, Figure 2 A schematic diagram of an implementation environment of the timeliness identification method of information provided in the embodiment of the present application is shown. In this implementation environment, the main hardware and software entities involved include an information channel end 210, a backend server 220, and a user end 230. Among them, the information channel end 210 is used to collect and integrate various types of information, process to form information-based products, and upload them to the backend server 220. For example, the information can be any one of news, advertisements, notifications, etc. The backend server 220 is used to push the information to the user end 230 after integration. The user end 230 can run related software or network applications with information receiving and display functions. For example, the software or network application can be a news browser, an online shopping program, a web page, or a small program that depends on the host program to run. In some embodiments, the information channel end 210 and the user end 230 can include any one or more of devices such as smart watches, smart phones, tablet computers, personal digital assistants (Personal Digital Assistant, PDA), laptops or desktop computers. The server 220 can be the backend server corresponding to the aforementioned software or network application. In some embodiments, the server 220 can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, or as a cloud server that provides cloud services, cloud databases, cloud computing, cloud storage, network services, and other services. In some embodiments, the information channel end 210 and the server 220, and the server 220 and the user end 230 can establish a communication connection through a wireless network or a wired network. The wireless network or wired network uses standard communication technology and / or protocols, and the network can be set to the Internet or any other network, such as but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network, or any combination of a virtual private network.

[0064] exist Figure 2In the implementation environment shown, the method provided by the present application can be executed in the server 220, that is, after the information is stored in the server 220 through the information channel terminal 210, the server 220 identifies the timeliness of the information by executing the timeliness identification method of the information in the embodiment of the present application, and pushes it to the user terminal 230 according to the result of the timeliness identification, for example, when the information expires, the push is stopped. In some embodiments, the method provided by the present application can also be executed in the user terminal 230, that is, after the information is stored in the server 220 through the information channel terminal 210, the server 220 pushes the information to the user terminal 230, and after the user terminal 230 receives the information, it identifies the timeliness of the information by executing the timeliness identification method of the information in the embodiment of the present application, thereby providing browsing guidance for the user according to the timeliness identification result, helping the user to read more valuable information and improving the user experience. Of course, it is understandable that Figure 2 The implementation environment in the embodiment of the present application is only an optional application scenario of the information time-limit identification method provided in the embodiment of the present application, and the actual application is not fixed. Figure 2 The hardware and software environment shown.

[0065] Reference Figure 3 , Figure 3 is an optional flow chart of the method for identifying the timeliness of information provided in the embodiments of the present application. Figure 3 The method mainly includes steps 310 to 340.

[0066] Step 310: Obtain target text data from the information to be identified;

[0067] In the embodiment of the present application, the information to be identified refers to the information whose timeliness needs to be identified. Here, timeliness refers to the effective period of the information. The data type of the information can be of various types, for example, it can be any type of information such as text, image, audio, etc. For information of different data types, the method of obtaining the target text data from it can also be adaptively selected. For example, referring to Figure 4When the information to be identified 410 itself is text-type data, it can be used as initial text data 420; when the information to be identified is audio-type data, automatic speech recognition technology (Automatic Speech Recognition, ASR) can be used to perform speech recognition on the audio data and convert it into text-type data; when the information to be identified is image-type data, OCR (Optical Character Recognition, optical character recognition) technology can be used to perform text recognition on the image data to obtain text-type data; similarly, when the information to be identified is video-type data, speech recognition can be performed on the audio data in the video data or text recognition can be performed on each frame of the video data to obtain text-type data. In this application, the text data converted from audio data, image data or video data can convert the content of the original data type information into text form to facilitate subsequent natural language processing. This part of the text data basically retains the content of the original information and can also be used as the initial text data 420.

[0068] In some embodiments, the initial text data 420 obtained by collection or conversion can be used as target text data, that is, the entire initial text data 420 is used as data for subsequent natural language processing; in some embodiments, part of the text content can also be selected from the initial text data 420 as target text data to reduce the amount of data processing and improve recognition efficiency. Specifically, when selecting part of the text content, a part of the natural language segment can be intercepted from at least one of the beginning or end of the initial text data 420 to obtain the target text data, for example, the first paragraph and the last paragraph of the initial text data 420 can be intercepted as the target text data; in some embodiments, it can also be intercepted according to the number of words or the proportion of the length, for example, the first quarter of the text in the initial text data 420 can be intercepted as the target text data; in some embodiments, sentences or paragraphs can also be randomly extracted from the initial text data 420 to obtain the target text data. For example, with Figure 5 For example, after obtaining the initial text data 420, the target text data 421 can be randomly extracted from the initial text data 420, for example Figure 5 The target text data 421 obtained in the above method includes the text content of "Study hard tomorrow, there will be an exam next week".

[0069] Here, it should be noted that for combined information containing multiple data types, the above-mentioned method can be used to extract the initial text data one by one for information of different data types, and then all the initial text data can be integrated and the integrated text data can be used as the target text data, or part of the text content can be selected from the integrated text data as the target text data; it is also possible to select information of one data type from the combined information and execute the recognition method in the embodiment of the present application alone, or to group the information according to different data types and execute the recognition method in the embodiment of the present application one by one, and integrate the multiple recognition results obtained.

[0070] Step 320: Segment the target text data to obtain a plurality of sentences, and determine first sentence feature data corresponding to each sentence;

[0071] In an embodiment of the present application, the acquired target text data is segmented at the sentence level, i.e., sentence segmentation, to obtain multiple sentences. The purpose of the sentence segmentation process here is to segment a longer text content into multiple shorter sentences, so as to analyze the semantics of the sentences through natural language processing technology and improve the recognition accuracy. Specifically, in some embodiments, the sentence segmentation process can be segmented based on the spacing characters in the target text data. For example, whenever a spacing character such as a comma, period, exclamation mark, or question mark appears in the target text data, the target text data is truncated at the spacing character, thereby splitting it into multiple sentences. For example, with Figure 5 For example, the target text data 421 in the example contains a comma, so the target text data "Study hard tomorrow, and have an exam next week" can be truncated and segmented into two sentences at the comma, which are recorded as sentence S1: "Study hard tomorrow" and sentence S2: "There will be an exam next week". Of course, in some text data, the spacing characters may have other functions and do not necessarily represent the separation of sentences. For example, the decimal point in mathematical characters and the period in English characters may be confused. For these more special spacing characters, a targeted decision tree can be established based on experience, and hierarchical judgments can be made according to empirical rules, thereby improving the accuracy of sentence segmentation processing.

[0072] For the sentences obtained by segmentation, the sentence feature data corresponding to these sentences can be further determined, which is recorded as the first sentence feature data. In the present application, sentence feature data is a data form used to characterize sentences. Since text-type data is unstructured data, it is difficult to be directly processed by a machine learning model, so it needs to be converted into structured sentence feature data. The data structure of the sentence feature data in the embodiment of the present application can be any one of a number, a vector, a matrix or a tensor.

[0073] The following describes in detail some methods for determining sentence feature data in embodiments of the present application:

[0074] First of all, it can be understood that the sentence itself is composed of words, and words are the most basic units in natural language processing technology. Therefore, in the embodiment of the present application, the sentence feature data used to characterize the sentence can be determined according to the words in each sentence. In some embodiments, the sentence can be first segmented to obtain multiple words that constitute the sentence, and these words are recorded as a phrase, and each sentence corresponds to a phrase. For example, for sentence S1, the text content it contains is "study hard tomorrow", which can be divided into three words "tomorrow", "well", and "study", so the phrase corresponding to sentence S1 can be recorded as phrase C1: {tomorrow, well, study}; similarly, the phrase corresponding to sentence S2 can be recorded as phrase C2: {next week, to, exam}. Specifically, there are many word segmentation algorithms that can be used in the embodiments of the present application. For example, in some embodiments, a word segmentation algorithm based on a dictionary can be used to first divide the sentence into words according to the dictionary, and then find the best combination of words; in some embodiments, a word segmentation algorithm based on characters can also be used to first divide the sentence into characters, and then combine the characters into words to find the best combination. The following uses a dictionary-based statistical language model as an example to explain and illustrate the principle and process of word segmentation.

[0075] In the concept of statistical language model, a natural language sentence consists of a series of words arranged in a specific order: for example, S is a sentence in natural language, S = {ω1, ω2, ω3, ..., ω n}, where {ω1,ω2,ω3,...,ω n} represents a word sequence, ω1, ω2, ω3...ω n Each represents a word, and n is a positive integer. Assume that sentence S can be divided into five words, represented by S = {ω1, ω2, ω3, ω4, ω5}. Then for sentence S, the statistical language model can determine the possibility of its appearance in all natural language sentences, that is, determine the probability of the appearance of sentence S, and record this probability as P(S). The probability P(S) can be expressed as P(S) = P(ω1, ω2, ω3, ω4, ω5). According to the relationship between conditional probabilities, we can know that:

[0076] P(S)=P(ω1)*P(ω2|ω1)*P(ω3|ω1,ω2)*P(ω4|ω1,ω2,ω3)*P(ω5|ω1,ω2,ω3,ω4)

[0077] In the formula, P(ω1) represents the probability of word ω1 appearing; P(ω2|ω1) represents the probability that the word following ω1 is ω2 when word ω1 appears; P(ω3|ω1,ω2) represents the probability that the word following ω2 is ω3 when words ω1 and ω2 appear consecutively; similarly, P(ω4|ω1,ω2,ω3) represents the probability that the word following ω3 is ω4 when words ω1, ω2 and ω3 appear consecutively; P(ω5|ω1,ω2,ω3,ω4) represents the probability that the word following ω4 is ω5 when words ω1, ω2, ω3 and ω4 appear consecutively.

[0078] Generally speaking, as the number of words in a word sequence increases, the conditional probability of the later words becoming more difficult to calculate. When there are more words in a word sequence, the correlation between two words that are not continuous in the word sequence may be weak. Therefore, we can use the Markov hypothesis and assume that the probability of any word appearing is only related to the word before it. For the sentence S mentioned above, the probability of its appearance can be determined as:

[0079] P(S)=P(ω1)*P(ω2|ω1)*P(ω3|ω2)*P(ω4|ω3)*P(ω5|ω4)

[0080] Where P(ω3|ω2) represents the probability that the word following ω2 is ω3 when ω2 appears; P(ω4|ω3) represents the probability that the word following ω3 is ω4 when ω3 appears; P(ω5|ω4) represents the probability that the word following ω4 is ω5 when ω4 appears.

[0081] In the aforementioned Markov hypothesis, it is assumed that the probability of occurrence of any word is only related to the word preceding it. In this case, the statistical language model for calculating the probability of occurrence of statement S is called a bigram model. In practical applications, it can be assumed that the probability of occurrence of any word is only related to the M words preceding it (M is a positive integer). The statistical language model for calculating the probability of occurrence of statement S obtained based on this assumption is called an M+1 bigram model. It can be understood that the statistical language models obtained under these assumptions can all be used to calculate the probability of occurrence of statement S. In the embodiments of the present application, for the convenience of explanation, the bigram model is used as an example for illustration.

[0082] When calculating the probability P(S) of the occurrence of statement S through a binary model, the results of P(ω1), P(ω2|ω1), P(ω3|ω2), P(ω4|ω3), and P(ω5|ω4) should be determined separately. According to the law of large numbers, as long as the amount of statistical data is sufficient, the relative frequency can be considered equal to the occurrence frequency. Therefore, P(S) can be determined through the following process: Select a corpus, determine the number of occurrences of word ω1, divide it by the size of the number of words in the corpus to obtain P(ω1); for P(ω2|ω1), determine the number of times word ω2 appears immediately after word ω1, and then divide it by the number of occurrences of word ω1 to obtain P(ω2|ω1). Similarly, the results of P(ω3|ω2), P(ω4|ω3), and P(ω5|ω4) can be obtained. Then, multiply P(ω1), P(ω2|ω1), P(ω3|ω2), P(ω4|ω3), and P(ω5|ω4) to obtain the probability P(S) of the statement S segmented as S = {ω1, ω2, ω3, ω4, ω5}. However, there may be multiple ways to divide the statement S into a sequence of 5 words, so it is necessary to determine the most accurate word segmentation result from them. Refer to Figure 6 , for example, taking the text content of "What he said is indeed reasonable" as an example, according to normal natural language logic, it can be divided into a word sequence: "he, said, of, indeed, reasonable". However, since "truly" and "really" are also frequently occurring words, directly dividing the statement according to the dictionary may also result in a word segmentation of "he, said, truly, really, reason". Among the above word segmentation results of the statement "What he said is indeed reasonable", they all contain 5 words in total, but the last three words in the sequence are different. Therefore, it is necessary to further distinguish the rationality of multiple word segmentation results and select the better one as the word segmentation result of the statement. Specifically, the first probability P1 of the statement S segmented as {he, said, of, indeed, reasonable} and the second probability P2 of the statement S segmented as {he, said, truly, really, reason}, as well as other possible word segmentation results in the form, such as Figure 6 the third probability P3 of the corresponding statement S segmented as {he, said, of, truly, really, reason} in Figure 6 the example shown. For more accuracy, the context of the statement S is also taken into account, that is, considering the breaks in the statement before the word "he" and after the word "reason", in <s>Indicates the separation of sentences, P(he| <s>) represents the probability that "he" is the first word of the sentence, P( <s>|The probability that the character "理" indicates that it is the last word in the sentence. So the first probability P1 = P(他| <s>)*P(say|he)*P(of|said)*P(indeed|of)*P(reason|indeed), similarly, the second probability P2 and the third probability P3 can be obtained. Comparing the calculated first probability P1, second probability P2 and third probability P3, it can be found that the first probability P1 is much larger than the probabilities corresponding to other word segmentation results, so it can be confirmed that the word segmentation result of sentence S should be {he, said, of, indeed, indeed}.

[0083] In the embodiment of the present application, after obtaining the phrase corresponding to each sentence, the first sentence feature data corresponding to the sentence can be determined according to the word feature data corresponding to each word in the phrase. In the present application, similarly, the word feature data is a data form used to characterize the word, and its data structure can also be any one of a numerical value, a vector, a matrix, or a tensor. For example, in some embodiments, each word can be represented by a numerical value, such as the Chinese word "I" can be represented by the numerical value "1"; in some embodiments, a dictionary can be established for all the words in the corpus, and then each word in the dictionary can be represented by a one-hot vector. For example, if there are 1000 words in the dictionary, the dimension of the one-hot vector is 1000 dimensions, and each word can be assigned a corresponding one-hot vector according to the order in which each word appears in the dictionary. For example, if "I" is the first word in the dictionary, the one-hot vector corresponding to "I" can be represented as (1, 0, 0...0, 0), where the first element in the vector is 1 and the remaining 999 elements are all 0; in some embodiments, in order to reduce the dimension of the vector representing the word, a word embedding vector can also be used to represent the word. The word embedding vector can be obtained by mapping the words in the dictionary to a vector space with a unified lower dimension. Strategies for generating such mappings include neural networks, dimensionality reduction of word co-occurrence matrices, probability models, and interpretable knowledge base methods. For example, taking the use of a neural network to obtain a word embedding vector as an example, the trained word2vec (word to vector) model can be used to map the word embedding vectors of each word in the dictionary. Compared with the aforementioned unique hot vector, the word embedding vector also has a certain semantic representativeness, that is, semantically similar words will also be closer in the vector space, so it is more widely used. Of course, the above implementation method is only used to illustrate the method of determining the word feature data in this application, and the actual implementation process is not limited to the above specific examples. For the convenience of explanation, the following uses the word embedding vector as the word feature data as an example to illustrate the calculation process of the first sentence feature data in this application.

[0084] Still taking the aforementioned sentence S1 as an example, its corresponding phrase is C1: {tomorrow, good, study}. Assuming that the word embedding vector corresponding to the word "tomorrow" in the phrase C1 is (0, 1, 0, 1), the word embedding vector corresponding to the word "good" is (1, 1, 3, 0), and the word embedding vector corresponding to the word "study" is (0, 0, 1, 2), then the word embedding vectors corresponding to each word in the phrase C1 can be accumulated, that is, the elements in the same position in the three word embedding vectors are summed, and the obtained elements form a new vector, which is recorded as the first vector. The first vector obtained by accumulating the word embedding vectors corresponding to the words "tomorrow", "good" and "study" in the phrase C1 is (1 , 2, 4, 3), in some embodiments, the first vector can be used as the first sentence feature data; in the present application, for the convenience of standardization and data processing, the first vector can be normalized and used as the first sentence feature data. Specifically, the normalization method here can be to limit the modulus of the first vector to a specific value, such as proportionally reducing each element in the first vector so that the modulus of the first vector is 1; or it can be to limit the sum of each element in the first vector to a specific value, such as proportionally reducing each element in the first vector so that the sum of each element in the first vector is 1, then the first vector will be adjusted to (0.1, 0.2, 0.4, 0.3).

[0085] The above-mentioned embodiments illustrate the specific process of determining the sentence feature data corresponding to a sentence from the words that make up the sentence. In some embodiments, for relatively short sentences, a sentence feature database can also be established to store the sentence feature data determined by the above method or other rules, and determine the association between the sentence feature data and the sentence. When the sentence is subsequently segmented from the text data, the sentence feature data corresponding to the sentence can be directly determined based on the association.

[0086] Step 330: input the first sentence feature data corresponding to each sentence into the timeliness recognition model to obtain the first timeliness recognition result corresponding to each sentence; the first timeliness recognition result is used to characterize the effective time of the sentence;

[0087] In the embodiment of the present application, after determining the first sentence feature data corresponding to each sentence, the first sentence feature data can be input into the trained time recognition model to obtain the time recognition result predicted by the time recognition model, and the time recognition result here is recorded as the first time recognition result, and each sentence involved in the prediction corresponds to a first time recognition result. In the present application, the first time recognition result is used to characterize the effective duration of the sentence predicted by the model. Here, the effective duration of the sentence refers to the duration of the content of the sentence maintaining the original meaning at the time level starting from the reference time node. Specifically, in the embodiment of the present application, the reference time node can generally be defaulted to the current moment, that is, the time node for time recognition of the information. Of course, in some embodiments, the reference time node can also be set to other time nodes, such as the generation time of the information, the collection time of the information, or the release time of the information. It can be understood that the reference time nodes corresponding to each sentence obtained by splitting the same information are the same, and are all determined by the reference time nodes corresponding to the information to which they belong. And in the embodiment of the present application, the relevant data of the reference time node can be pre-input into the model as a hyperparameter, so that the model predicts the first time recognition result of the sentence based on the set reference time node. Specifically, the type of the first aging recognition result can be flexibly set as needed, and can be achieved by selecting a machine learning algorithm that performs a corresponding task to train an aging recognition model, or by adjusting the hyperparameters of the aging recognition model.

[0088] For example, in some embodiments, the timeliness identification in the present application can be set as a classification task, and the classification task can be performed by machine learning algorithms such as SVM classifier, K nearest neighbor algorithm, decision tree algorithm, naive Bayes algorithm, etc. Specifically, different timeliness categories can be pre-defined, for example, the timeliness greater than 3 days is defined as the "long timeliness" category, and the timeliness less than or equal to 3 days is defined as the "short timeliness" category. Taking statements S1 and S2 as examples, assuming that the day the information is produced is Monday, the date is March 1, then for statement S1: "Study hard tomorrow", the "tomorrow" is for tomorrow, March 1, that is, March 2, so the timeliness of statement S1 is valid on March 1. If the statement S1 is released on March 2 or later, the word "tomorrow" will be misleading and the information will lose its original meaning. Therefore, for statement S1, its timeliness belongs to the category of less than or equal to 3 days, and the first timeliness identification result corresponding to statement S1 is the "short timeliness" category. As for the statement S2: "There will be an exam next week", "next week" refers to the week after March 1 to March 7. Therefore, if the statement S2 is released on any day between March 1 and March 7, the reference to "next week" is clear and unambiguous. When the release period exceeds 7 days, that is, if the statement S2 is released on March 8 or later, the word "next week" will be misleading and the information will lose its original meaning. Similarly, for statement S2, its validity period includes the time period from March 1 to March 7, which is as long as seven days and belongs to the category of more than 3 days. At this time, the first validity period recognition result corresponding to statement S2 is the "long validity period" category. Of course, the above-mentioned machine learning algorithm and the classification category setting method of validity are only used to illustrate the execution principle of validity recognition under the classification task. In actual application, the type of machine learning algorithm, the classification category setting method and the number of validity can be flexibly set according to needs, and will not be elaborated here.

[0089] In some embodiments, the timeliness identification in this application can also be set as a regression task, that is, a specific timeliness value is predicted by machine learning algorithms such as linear regression, logistic regression, gradient boosting decision tree (GBDT), extreme gradient boosting (eXtreme Gradient Boosting, XGBoost), etc., and the prediction accuracy can be set to any time scale, for example, it can be hours, days, weeks, etc. Still taking statements S1 and S2 as examples, assuming that the information is produced on Monday and the date is March 1, the prediction accuracy of the machine learning model is based on the number of days. Similarly, for statement S1: "Study hard tomorrow", it is valid on March 1, so the first timeliness identification result corresponding to statement S1 is 1 day. For statement S2: "There will be an exam next week", it is valid from March 1 to March 7, so the first timeliness identification result corresponding to statement S2 is 7 days. Similarly, in the actual application of this embodiment, the type of machine learning algorithm and the prediction accuracy of timeliness can be flexibly set as needed.

[0090] Step 340: Determine a second timeliness identification result corresponding to the information according to the first timeliness identification result that represents a shorter effective time; the second timeliness identification result is used to represent the effective time of the information.

[0091] In an embodiment of the present application, after determining the first timeliness recognition result corresponding to each sentence in the target text data, the timeliness recognition result of the information to which the target text data belongs can be determined based on the first timeliness recognition result corresponding to the sentence. The timeliness recognition result of the information is recorded as the second timeliness recognition result, and the second timeliness recognition result is used to characterize the effective duration of the information. Here, the effective duration of the information refers to the duration during which the content of the information maintains its original meaning at the time level starting from the reference time node. As mentioned above, the reference time nodes corresponding to the various sentences obtained by splitting the same information are the same, and are all determined by the reference time nodes corresponding to the information to which they belong, that is, the starting measurement time nodes of the effective duration of the information and the sentence are consistent. It is understandable that since each sentence is extracted from the information, each sentence can represent the timeliness of the information to a certain extent, and the actual timeliness of the information should be determined based on the sentence with a shorter effective time among all the sentences. The reason is that for sentences with a shorter effective time, once their time limit is exceeded, the content of the sentence will not be able to maintain the original meaning at the time level, and the sentence itself will be misleading. Even if other sentences with a longer effective time can still correctly provide some information guidance in the future, from the information dimension composed of all the sentences, there are still problems of misleading and unclearness as a whole. Therefore, the timeliness of the information should take the shorter effective time among all the sentences, that is, the second timeliness recognition result is determined based on the first timeliness recognition result with a shorter effective time.

[0092] For example, refer to Figure 7 , taking the information containing sentence S1 and sentence S2 as an example, assuming that the day the information is produced is Monday, the date is March 1st, the first timeliness recognition result corresponding to sentence S1 is 1 day, and the first timeliness recognition result corresponding to sentence S2 is 7 days. However, after March 1st, the "tomorrow" in the information will lose its original meaning of March 2nd, that is, the meaning of "study hard on March 2nd" originally contained in the information will be distorted, making the entire information lose its value. Therefore, the second timeliness recognition result corresponding to the information should be the first timeliness recognition result of sentence S1, that is, the timeliness of the information is 1 day, and it is only valid before March 1st. Similarly, when the first time limit identification result represents the classification result of the time limit length interval, the above method can also be used to determine the second time limit identification result corresponding to the information. For example, the first time limit identification result corresponding to sentence S2 is the "long time limit" category, and the first time limit identification result corresponding to sentence S1 is the "short time limit" category. The second time limit identification result corresponding to the actual information should be the first time limit identification result with a shorter effective time represented, that is, the second time limit identification result is also in the "short time limit" category. The above takes two sentences as an example to illustrate the process of determining the second time limit identification result corresponding to the information from the first time limit identification results corresponding to multiple sentences in this application. It can be understood that when the number of sentences exceeds two, the first time limit identification result with a shorter effective time represented can still be selected as the second time limit identification result corresponding to the information, which will not be elaborated here.

[0093] The aforementioned embodiments of the present application explain the scheme for identifying and predicting the timeliness of information at the sentence level, while in other embodiments, the timeliness of information can also be determined by semantic matching.

[0094] Specifically, some corpus matching libraries can be pre-established, and the corpus matching libraries can include multiple time-sensitive keywords and time-sensitive key sentences. The content of the corpus matching library can be selected from existing corpora, for example, refer to Table 1:

[0095]

[0096] Table 1

[0097] Table 1 shows some commonly used corpora that contain time-limited information. For example, the corpora of the same day type belong to information that expires on the same day, such as "tonight" and "tomorrow morning", which will no longer be valid after today; the corpora of the same week type belong to information that expires on the same week, such as "this Wednesday" and "weekend", which will no longer be valid after this week. Therefore, the words and sentences in Table 1 can be selected as the basis for constructing a corpus matching library for this application, and the corpus matching library can be expanded through the similarity between words or sentences. Of course, the above corpora are simple examples, and there may be inaccuracies in the specific time limits. When applied to the recognition method of this application, other rules can be set to effectively correct them, or the selected corpus matching library content can be flexibly adjusted and deleted based on experience.

[0098] In an embodiment of the present application, the contents in the established corpus matching library are divided into time-sensitive keywords and time-sensitive key sentences. Here, time-sensitive keywords refer to words with time meanings, such as "tomorrow", "tonight", "two hours", "Friday", "three days later", "Spring Festival", "May Day", etc.; similarly, time-sensitive key sentences can be sentences with time meanings, such as "expected to be next week", "the competition will last for 21 days", "the event will end on March 6", "are you free next month", etc. Based on the corpus matching library, it can be matched with the target text data to obtain a matching result, so as to determine the timeliness of the information based on the matching result, and compare it with the obtained second timeliness recognition result, so as to update the second timeliness recognition result and improve the accuracy of timeliness recognition.

[0099] Specifically, when matching the target text data, in some embodiments, several words can be extracted from the target text data, and these words are matched with the time-effective keywords for similarity, and the similarity between the extracted words and the time-effective keywords is determined, which can be recorded as the first similarity, and then the first similarity is compared with a pre-set similarity threshold, which can be recorded as the first threshold. Taking the numerical value of the similarity as 0 to 1 as an example (when the similarity is 1, it means that the extracted word and the time-effective keyword are completely consistent), the size of the first threshold can be set to 0.8. When the first similarity between a certain time-effective keyword and the extracted word is greater than the first threshold, the matching result can be determined based on the time-effective keyword. For example, assuming that the words extracted from the target text data are "tomorrow", "well", and "learning", when the time-effective keyword "tomorrow" is matched, the first similarity between the extracted word "tomorrow" and the time-effective keyword "tomorrow" is 1, the first similarity between the extracted word "well" and the time-effective keyword "tomorrow" is 0.2, and the first similarity between the extracted word "learning" and the time-effective keyword "tomorrow" is 0.3. Then, the first similarity between the extracted word "tomorrow" and the time-effective keyword "tomorrow" is greater than the first threshold, and the time-effective keyword "tomorrow" can be used as a matching result. Similarly, in some embodiments, several sentences can also be extracted from the target text data, and these sentences are matched with the time-effective key sentences for similarity, and the similarity between the extracted sentence and the time-effective key sentence is determined, which can be recorded as the second similarity here, and then the second similarity is compared with a pre-set similarity threshold, and the similarity threshold here can be recorded as the second threshold. When the second similarity between a certain time-effective key sentence and the extracted sentence is greater than the second threshold, the matching result can be determined according to the time-effective key sentence. It should be noted that the values ​​of the first threshold and the second threshold can be flexibly set as needed and are not fixed to the specific values ​​given in the example.

[0100] In the embodiment of the present application, when determining the first similarity between the extracted word and the time-effective keyword and determining the second similarity between the extracted sentence and the time-effective keyword sentence, the similarity can be obtained by calculating the feature data of the two. Specifically, taking the calculation of the first similarity as an example, the word feature data of the extracted word can be first determined, recorded as the first word feature data, and the word feature data of the time-effective keyword can be determined, recorded as the second word feature data, and then the difference value between the first word feature data and the second word feature data can be evaluated. Specifically, for example, when the word feature data are all in the form of numerical values, the difference between the numerical values ​​of the first word feature data and the second word feature data can be calculated, and the difference can be used as the difference value between the two. Similarly, the difference value can also be determined by the spatial distance between vectors, the norm difference between matrices, etc., where the spatial distance between vectors can be determined by Euclidean distance, Manhattan distance or Hamming distance, etc. In the embodiment of the present application, the corresponding calculation method can be selected according to the data structure of the word feature data. It can be understood that there is a negative correlation between the first similarity and the difference value, that is, the larger the difference value, the smaller the first similarity, and vice versa, the smaller the difference value, the larger the first similarity. Therefore, the first similarity can be further determined based on the obtained difference value, for example, by an inverse proportional function.

[0101] In some embodiments, the first similarity and the second similarity can also be directly determined by some vector-based similarity algorithms. For example, taking the first similarity as an example, the vector of the extracted word and the vector of the time-effective keyword can be determined. For example, both the extracted word and the time-effective keyword are represented by word embedding vectors, and the vector corresponding to the extracted word is recorded as the first word embedding vector, and the vector corresponding to the time-effective keyword is recorded as the second word embedding vector. Then, the first similarity can be calculated based on the first word embedding vector and the second word embedding vector by algorithms such as the cosine similarity algorithm, the Pearson correlation coefficient method, or the Jaccard similarity coefficient method. Specifically, for example, the length of the first word embedding vector can be first determined, recorded as the first length, and the length of the second word embedding vector can be recorded as the second length. Then, the product of the first length and the second length is calculated as the first value, and the inner product of the first vector and the second vector is calculated as the second value, and then the quotient of the first value and the second value is calculated as the first similarity between the first word embedding vector and the second word embedding vector, that is, the first similarity between the extracted word and the time-effective keyword. The algorithm formula used in the embodiment of the present application is as follows:

[0102]

[0103] In the formula, a represents the extracted word, b represents the time-sensitive keyword; sim(a,b) represents the first similarity; Represents the first word embedding vector corresponding to the extracted word; represents the second word embedding vector corresponding to the time-sensitive keyword; "·" represents the inner product between vectors; Represents the length of the first word embedding vector; Represents the length of the second word embedding vector.

[0104] It is understandable that in the embodiment of the present application, the above-mentioned similarity calculation method can also be used to determine the second similarity between the extracted sentence and the time-limited key sentence, and the specific process will not be repeated here. In the embodiment of the present application, according to the similarity and the pre-set threshold, the matching result of the time-limited keyword or time-limited sentence can be obtained, and the form of the matching result can generally be divided into two types: the first is that the time-limited keyword or time-limited sentence contains a time-limited period, such as "two hours", "three days later", "the game will last for 21 days", etc., from which the length of the time limit can be directly obtained; the second is that the time-limited keyword or time-limited sentence contains a specific time node, such as "tomorrow", "tonight", "Spring Festival", "expected to be next week", "the event will end on March 6", etc. For this type of matching result, the time-limited recognition result of the information can be determined based on the time node contained therein and the current time, and the time-limited recognition result here is recorded as the third time-limited recognition result. For example, for the time-limited keyword "tomorrow", the time limit relative to the current time is 1 day; assuming that today is March 1, for the time-limited key sentence "the event will end on March 6", the time limit relative to March 1 is 5 days. Of course, it is understandable that if the form of the time limit recognition result is preset as the aforementioned "long time limit" and "short time limit" categories, the category to which the third time limit recognition result belongs can also be determined according to the length of the obtained time limit.

[0105] It should be noted that when there are multiple matching results in the matching process, the corresponding third time-limited identification result can be determined for each matching result first, and then the obtained third time-limited identification result can be screened, and the third time-limited identification result represented by the shorter effective time length is retained as the third time-limited identification result finally determined according to the matching result. The principle of screening here is similar to the aforementioned step 340, which will not be repeated here. In addition, in an embodiment of the present application, when the third time-limited identification result is determined according to the matching result, the second time-limited identification result corresponding to the information can be updated, that is, the shorter effective time length represented in the third time-limited identification result and the second time-limited identification result is determined as the updated second time-limited identification result. For example, according to the aforementioned method of prediction through a machine learning model, the second time-limited identification result corresponding to the information is identified as 3 days, and through the method of matching the corpus matching library, the third time-limited identification result corresponding to the information is identified as 1 day, then it is obvious that the more accurate time-limited of the information is 1 day, so the second time-limited identification result of the updated information is 1 day.

[0106] The following is a description of an implementation process of the method for identifying the timeliness of information in the embodiment of the present application in conjunction with a specific embodiment.

[0107] Reference Figure 8 In the embodiment of the present application, after obtaining the information that needs to be time-limited, the information can be predicted based on the time-limited recognition model and matched based on the corpus matching library, thereby obtaining the second time-limited recognition result output by the time-limited recognition model and the third time-limited recognition result output by the corpus matching library. Specifically, the target text data can be firstly extracted from the information, and then for the target text data, a number of words can be randomly extracted from it, such as Figure 8 m words can be extracted from the text, which can be recorded as word 1, word 2, ..., word m. On the other hand, the target text data can also be divided into multiple sentences, for example Figure 8 The code is split into n statements, which can be recorded as statement 1, statement 2, ..., statement n. Here, m and n are both positive integers.

[0108] After obtaining the segmented sentences, they can be converted into sentence feature data. The sentence feature data obtained by converting sentence 1 can be recorded as sentence feature data 1. Similarly, the sentence feature data obtained by converting sentence n can be recorded as sentence feature data n. The total number of sentence feature data is also n. After inputting the n sentence feature data into the timeliness recognition model one by one, the n first timeliness recognition results output by the model can be obtained ( Figure 8 (not shown), and according to the n first aging recognition results, the second aging recognition result based on the prediction output of the aging recognition model is determined through the aforementioned step 340.

[0109] The m words and n sentences obtained can be input together into the corpus matching library for matching. The specific matching method has been described in the previous embodiment and will not be repeated here. The result output after matching is recorded as the third timeliness recognition result. The third timeliness result obtained by matching can be taken into comprehensive consideration to update the second timeliness recognition result based on the prediction output of the timeliness recognition model, so as to obtain a more accurate timeliness recognition result of the information. Of course, it should be noted here that in the timeliness recognition method of information provided in this application, that is, it can be based on Figure 8 In the above method, the timeliness identification model and the corpus matching library are used in combination, or one of the methods can be used alone to identify the timeliness of information. For example, the corpus matching work can be performed alone, and the third timeliness identification result can be used as the timeliness identification result of the information. Of course, it is also possible to remove Figure 8 In the processing flow of the middle part, for example, only the extracted words can be input into the corpus matching library, which is not limited in the embodiments of the present application.

[0110] Reference Fig. 9 In an embodiment of the present application, a method for training an information timeliness recognition model is also provided. Similarly, the information timeliness recognition model training can be applied in a terminal, in a server, or in software in a terminal or a server to implement a portion of the software functions. Fig. 9 FIG. 9 is an optional flow chart of a method for training a time-effect recognition model of information provided in an embodiment of the present application. The method mainly includes steps 910 to 950:

[0111] Step 910: Obtain batches of sample sentences and time-limit tags corresponding to the sample sentences; the time-limit tags are used to represent the effective time of the sample sentences;

[0112] In the embodiment of the present application, when training the timeliness recognition model, the corresponding machine learning algorithm can be selected according to the needs of the task, and the hyperparameters of the timeliness recognition model can be adjusted. For example, when it is desired to determine whether the timeliness of the information is within a predetermined timeliness interval, a machine learning algorithm that is convenient for performing classification tasks can be selected, and the number of categories of the classification output can be adjusted to achieve the purpose of setting the number of timeliness intervals; when it is desired to determine the specific duration of the timeliness of the information, a machine learning algorithm that is convenient for performing regression tasks can be selected, and the numerical granularity of the output can be adjusted to achieve the purpose of setting the prediction accuracy. In the specific training process, batches of sample sentences can be obtained, and these sample sentences can come from a variety of information. Then, according to the timeliness corresponding to the sample sentences, the timeliness labels corresponding to the sample sentences are marked. The timeliness labels are used to characterize the effective duration of the sample sentences, and can be category labels or numerical labels. In the embodiment of the present application, the training process of the model is explained by taking the classification task as an example. For example, two classification categories of "long validity" and "short validity" can be set. When the effective duration of a sentence is greater than 3 days, it is classified as the "long validity" category. When the effective duration of a sentence is less than or equal to 3 days, it is classified as the "short validity" category. The validity label corresponding to the "long validity" category can be set to a value of 1, and the validity label corresponding to the "short validity" category can be set to a value of 0. Here, the setting method for the reference time node of the effective duration corresponding to the sample sentence is similar to the description in step 330, and will not be repeated here.

[0113] Step 920: determine the second sentence feature data of each sample sentence;

[0114] In the embodiment of the present application, for each sample sentence, the sentence feature data corresponding to the sample sentence can be determined and recorded as the second sentence feature data. Specifically, the sentence feature data can be determined by referring to the aforementioned step 220.

[0115] Step 930: input the second sentence feature data into the time-limit recognition model to obtain a fourth time-limit recognition result of the sample sentence; the fourth time-limit recognition result is used to characterize the effective time of the sample sentence;

[0116] Step 940: Determine the loss value of training according to the fourth timeliness recognition result and the timeliness label;

[0117] Step 950: Update the parameters of the aging recognition model according to the loss value to obtain a trained aging recognition model.

[0118] In the embodiment of the present application, the second sentence feature data corresponding to the sample sentence is input into the timeliness recognition model to obtain the prediction results corresponding to each sample sentence, which are recorded as the fourth timeliness recognition result. The fourth timeliness recognition result is used to characterize whether the timeliness of the sample sentence belongs to the "short timeliness" category or the "long timeliness" category. For example, when the value of the output fourth timeliness recognition result is 0, it indicates that the timeliness of the sample sentence predicted by the timeliness recognition model belongs to the "short timeliness" category, and the timeliness of the sample sentence is less than or equal to 3 days; when the value of the output fourth timeliness recognition result is 1, it indicates that the timeliness of the sample sentence predicted by the timeliness recognition model belongs to the "long timeliness" category, and the timeliness of the sample sentence is greater than 3 days. Through the fourth timeliness recognition result and timeliness label corresponding to each sample sentence, the loss value predicted by the timeliness recognition model for each sample sentence can be determined. In the embodiment of the present application, the loss value can be determined by a 0-1 loss function. When the time recognition model predicts correctly, that is, when the fourth time recognition result corresponding to the sample sentence is consistent with the time label, the loss value is recorded as 0; when the time recognition model predicts incorrectly, that is, when the fourth time recognition result corresponding to the sample sentence is inconsistent with the time label, the loss value is recorded as 1. Of course, the actual loss value calculation method that can be used is not limited to the above-mentioned loss function. For example, the square difference loss function, the cross entropy loss function, etc. are all optional types. By minimizing the sum of the loss values ​​of multiple sample sentences predicted by the constrained time recognition model, the time recognition model can be back-propagated and trained to update its internal parameters, thereby obtaining a trained time recognition model. Here, it should be noted that, since in actual text data, the vast majority of sentences may not cover content related to time, in the present application, a classification category of irrelevant sentences can be set during the training and application of the machine learning model, or the labels of these irrelevant sentences can be marked as the longest time label during the training process to reduce interference with the accuracy of the trained model and improve the accuracy of recognition.

[0119] Reference Fig.10 In an embodiment of the present application, a method for pushing information is also provided. Similarly, the method for pushing information can be applied to a terminal, a server, or software in a terminal or a server to implement a portion of a software function. Fig.10 FIG. 1 is an optional flow chart of a method for pushing information provided in an embodiment of the present application. The method mainly includes steps 1010 to 1030:

[0120] Step 1010: Obtain information to be pushed;

[0121] Step 1020: Figure 3 The method shown performs timeliness identification on the information and obtains a timeliness identification result;

[0122] Step 1030: Push information according to the timeliness identification result, or push information and the timeliness identification result.

[0123] In the embodiment of the present application, after obtaining the information to be pushed, the following method can be used: Figure 3 The timeliness identification method of the information shown identifies the timeliness of the information, thereby obtaining the timeliness identification result of the information. When pushing information, the information can be pushed according to the timeliness identification result corresponding to the information, so that users can browse it, reduce the situation of pushing expired information, and improve user experience. In some embodiments, information can be pushed according to the timeliness identification result. For example, if the timeliness identification result corresponding to a certain information is 3 days, and the reference time node corresponding to the timeliness identification result is the time of listing, then the removal date can be marked when pushing the information, and the information will be removed three days after it is listed. For example, referring to Fig.11 In a shopping week promotion held by a shopping mall, a certain type of household electrical appliances is sold at a half-price discount, and the promotion will last for one week. In this application scenario, the mall operator or cooperative platform can use the information push method provided in the embodiment of the present application when publishing promotional advertisements, and automatically remove the promotional advertisements before the end of this week to reduce the occurrence of possible disputes such as misleading and deceiving consumers. In some embodiments, the order in which information is pushed can also be determined based on the timeliness identification results. For example, information with a shorter timeliness can be pushed first, and then information with a longer timeliness can be pushed. This can reasonably arrange the push tasks of each information, which is conducive to improving the hardware utilization efficiency of the push system and reducing the situation of pushing expired information. In some embodiments, the information and the timeliness identification results corresponding to the information can also be recommended to the user at the same time, so that the user can clearly understand the time limit of the information, reduce possible misleading, and improve the accuracy and professionalism of the information recommendation. In addition, it is understandable that Figure 3 The contents of the information time-limit identification method embodiment shown in the figure are applicable to the information push method embodiment of this invention. The functions specifically implemented in the information push method embodiment in the process of identifying the time-limit of information are the same as those in the embodiment of the information push method of this invention. Figure 3 The information timeliness identification method embodiment shown is the same as that of the embodiment, and the beneficial effects achieved are the same as those of the embodiment shown in FIG. Figure 3 The beneficial effects achieved by the embodiment of the method for identifying the timeliness of information shown are also the same.

[0124] Reference Fig.12 The present application also discloses a device for identifying the timeliness of information, including:

[0125] An acquisition module 1210 is used to acquire target text data from the information to be identified;

[0126] The segmentation module 1220 is used to perform sentence segmentation processing on the target text data to obtain multiple sentences and determine the first sentence feature data corresponding to each sentence;

[0127] The recognition module 1230 is used to input the first sentence feature data corresponding to each sentence into the timeliness recognition model to obtain the first timeliness recognition result corresponding to each sentence; the first timeliness recognition result is used to characterize the effective time of the sentence;

[0128] The processing module 1240 is used to determine a second timeliness identification result corresponding to the information according to the first timeliness identification result which represents a shorter effective time; the second timeliness identification result is used to represent the effective time of the information.

[0129] Optionally, in some embodiments, the acquisition module includes:

[0130] An initial text data acquisition unit, used for acquiring initial text data from the information;

[0131] The target text data acquisition unit is used to obtain the target text data by intercepting at least one of the head or the tail of the initial text data, or by randomly extracting the target text data from the initial text data.

[0132] Optionally, in some embodiments, the initial text data acquisition unit includes at least one of the following:

[0133] An audio data processing unit, for performing speech recognition on the audio data to obtain initial text data when the information is audio data;

[0134] An image data processing unit, used for performing text recognition on the image data to obtain initial text data when the information is image data;

[0135] The video data processing unit is used to perform speech recognition or text recognition on the video data when the information is video data to obtain initial text data.

[0136] Optionally, in some embodiments, the segmentation module includes:

[0137] A word segmentation processing unit is used to perform word segmentation processing on the sentence, obtain a phrase corresponding to the sentence, and determine word feature data corresponding to each word in the phrase; wherein each phrase includes a plurality of words;

[0138] The feature processing unit is used to determine the first sentence feature data corresponding to the sentence according to the word feature data corresponding to each word in the phrase.

[0139] Optionally, in some embodiments, the word feature data includes a word embedding vector;

[0140] The word segmentation unit is used to:

[0141] Map the words to the vector space and get the word embedding vector corresponding to the word.

[0142] Optionally, in some embodiments, the feature processing unit includes:

[0143] An accumulation subunit, used for accumulating the word embedding vectors corresponding to each word in the phrase to obtain a first vector;

[0144] The normalization subunit is used to normalize the first vector to obtain first sentence feature data.

[0145] Optionally, in some embodiments, the device further comprises:

[0146] A matching unit is used to match the target text data according to a pre-established corpus matching library to obtain a matching result; the corpus matching library includes a plurality of time-sensitive keywords and time-sensitive key sentences;

[0147] An updating unit is used to update the second timeliness recognition result according to the matching result.

[0148] Optionally, in some embodiments, the matching unit includes:

[0149] A first extraction subunit is used to extract a number of words from the target text data;

[0150] A first similarity processing unit, used to determine a first similarity between the word and the time-sensitive keyword;

[0151] A first matching subunit, configured to determine a matching result according to a time-sensitive keyword corresponding to a first similarity greater than a first threshold;

[0152] or,

[0153] A second extraction subunit is used to extract a plurality of sentences from the target text data;

[0154] A second similarity processing unit, used to determine a second similarity between the sentence and the time-critical sentence;

[0155] The second matching subunit is used to determine a matching result according to the time-sensitive key sentence corresponding to the second similarity greater than a second threshold.

[0156] Optionally, in some embodiments, the first similarity processing unit includes:

[0157] An embedding vector processing subunit, used to determine a first word embedding vector corresponding to a word and a second word embedding vector corresponding to a time-sensitive keyword;

[0158] A first calculation subunit, configured to calculate a first length of the first word embedding vector and a second length of the second word embedding vector;

[0159] A second calculation subunit, configured to obtain a first value according to a product of the first length and the second length, and obtain a second value according to an inner product of the first word embedding vector and the second word embedding vector;

[0160] The third calculation subunit is used to obtain the first similarity according to the quotient of the second value and the first value.

[0161] Optionally, in some embodiments, the updating unit includes:

[0162] The timeliness identification subunit is used to determine a third timeliness identification result of the information according to the matching result and the current time; the third timeliness identification result is used to represent the effective time of the information;

[0163] The time limit updating subunit is used to determine the shorter effective time represented by the third time limit identification result and the second time limit identification result as the updated second time limit identification result.

[0164] Understandably, Figure 3 The contents of the information time-limit identification method embodiment shown in the figure are applicable to the information time-limit identification device embodiment of the present invention. The functions specifically implemented by the information time-limit identification device embodiment of the present invention are the same as those of the present invention. Figure 3 The information timeliness identification method embodiment shown is the same as that of the embodiment, and the beneficial effects achieved are the same as those of the embodiment shown in FIG. Figure 3 The beneficial effects achieved by the embodiment of the method for identifying the timeliness of information shown are also the same.

[0165] Reference Fig.13 , the embodiment of the present application further discloses a computer device, including:

[0166] at least one processor 1310;

[0167] At least one memory 1320, used to store at least one program;

[0168] When at least one program is executed by at least one processor 1310, the at least one processor 1310 implements the following Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 An embodiment of a method for pushing information is shown.

[0169] It is understandable that if Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The contents of the information push method embodiment shown in FIG. 1 are applicable to the computer device embodiment. The functions specifically implemented by the computer device embodiment are similar to those in FIG. Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The information push method embodiment shown is the same as that shown in the embodiment, and the beneficial effects achieved are the same as those shown in the embodiment. Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The beneficial effects achieved by the information push method embodiment shown are also the same.

[0170] The present application also discloses a computer-readable storage medium in which a program executable by a processor is stored. When the program executable by the processor is executed by the processor, it is used to implement the following Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 An embodiment of a method for pushing information is shown.

[0171] Understandably, Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The contents of the information push method embodiment shown in the figure are applicable to the computer-readable storage medium embodiment, and the functions specifically implemented by the computer-readable storage medium embodiment are the same as those in the embodiment of the present invention. Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The information push method embodiment shown is the same as that of the embodiment, and the beneficial effects achieved are the same as those of the embodiment shown in FIG. Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The beneficial effects achieved by the information push method embodiment shown are also the same.

[0172] The embodiment of the present application further discloses a computer program product or a computer program, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in the above-mentioned computer-readable storage medium; Fig.13 The processor of the computer device shown can read the computer instruction from the above-mentioned computer readable storage medium, and the processor executes the computer instruction so that the computer device executes Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 An embodiment of a method for pushing information is shown.

[0173] Understandably, Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The contents of the information push method embodiment shown in the figure are applicable to the present computer program product or computer program embodiment, and the functions specifically implemented by the present computer program product or computer program embodiment are the same as those in the present computer program product or computer program embodiment. Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The information push method embodiment shown is the same as that of the embodiment, and the beneficial effects achieved are the same as those of the embodiment shown in FIG. Figure 3 The information timeliness identification method embodiment shown, Fig. 9 The training method embodiment of the time-effect recognition model shown, Fig.10 The beneficial effects achieved by the information push method embodiment shown are also the same.

[0174] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the application is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part described as a larger operation is performed independently.

[0175] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present application. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional techniques of the engineer. Therefore, those skilled in the art can implement the present application set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is determined by the full scope of the attached claims and their equivalents.

[0176] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0177] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, apparatus or device and execute the instructions), or in conjunction with such instruction execution system, apparatus or device. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, apparatus or device, or in conjunction with such instruction execution system, apparatus or device.

[0178] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0179] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0180] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0181] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.

[0182] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Technical personnel familiar with the field can make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.< / s> < / s> < / s> < / s>

Claims

1. A method for identifying the timeliness of information, characterized in that: The following steps are involved: Obtaining initial text data from the information to be identified; intercepting the target text data from at least one of the header or the tail of the initial text data; Performing sentence segmentation processing on the target text data to obtain a plurality of sentences, and determining first sentence feature data corresponding to each of the sentences; Inputting the first sentence feature data corresponding to each of the sentences into a time-limited recognition model to obtain a first time-limited recognition result corresponding to each of the sentences; The first time validity recognition result is used to represent the validity period of the sentence; Determine a second timeliness identification result corresponding to the information according to the first timeliness identification result characterized by a shorter effective time; The second time validity identification result is used to represent the validity period of the information; The method further comprises the following steps: The target text data is matched according to a pre-established corpus matching library to obtain a matching result; the corpus matching library includes a plurality of time-sensitive key sentences; the matching result is obtained by matching the target text data according to the pre-established corpus matching library, including: determining a second similarity between the sentence and the time-sensitive key sentence, and determining the matching result according to the time-sensitive key sentence corresponding to the second similarity greater than a second threshold; Determine a third timeliness identification result of the information according to the matching result and the current time; the third timeliness identification result is used to characterize the validity period of the information; The third timeliness identification result and the second timeliness identification result having the shorter effective time is determined as the updated second timeliness identification result.

2. The method according to claim 1, characterized in that The method further comprises: The target text data is randomly extracted from the initial text data.

3. The method according to claim 2, characterized in that The step of obtaining initial text data from the information to be identified comprises at least one of the following steps: When the information is audio data, performing speech recognition on the audio data to obtain the initial text data; When the information is picture data, performing text recognition on the picture data to obtain the initial text data; When the information is video data, voice recognition or text recognition is performed on the video data to obtain the initial text data.

4. The method according to claim 1, characterized in that: The determining of the first sentence feature data corresponding to each of the sentences includes: Perform word segmentation processing on the sentence to obtain a phrase corresponding to the sentence, and determine word feature data corresponding to each word in the phrase; wherein each phrase includes a plurality of words; The first sentence feature data corresponding to the sentence is determined according to the word feature data corresponding to each word in the phrase.

5. The method according to claim 4, characterized in that The word feature data includes a word embedding vector; The determining of word feature data corresponding to each word in the phrase includes: The word is mapped to the vector space to obtain the word embedding vector corresponding to the word.

6. The method according to claim 5, characterized in that The determining, based on the word feature data corresponding to each word in the phrase, first sentence feature data corresponding to the sentence includes: Accumulating the word embedding vectors corresponding to each word in the phrase to obtain a first vector; The first vector is normalized to obtain the first sentence feature data.

7. The method according to claim 1, characterized in that The corpus matching library also includes a plurality of time-sensitive keywords; the matching of the target text data to obtain a matching result also includes: Extracting a number of words from the target text data; Determining a first similarity between the word and the time-sensitive keyword; The matching result is determined according to the time-effective keyword corresponding to the first similarity greater than a first threshold.

8. The method according to claim 7, characterized in that The determining of the first similarity between the word and the time-effective keyword includes: Determine a first word embedding vector corresponding to the word and a second word embedding vector corresponding to the time-sensitive keyword; Calculate a first length of the first word embedding vector and a second length of the second word embedding vector; Obtain a first value according to the product of the first length and the second length, and obtain a second value according to the inner product of the first word embedding vector and the second word embedding vector; The first similarity is obtained according to the quotient of the second value and the first value.

9. A method for pushing information, characterized in that: The following steps are involved: Get the information to be pushed; Performing timeliness identification on the information by the method according to any one of claims 1 to 8 to obtain a timeliness identification result; Pushing the information according to the timeliness identification result; Or push the information and the timeliness identification result.

10. A device for identifying the timeliness of information, characterized in that: include: An acquisition module, used for acquiring initial text data from the information to be identified; intercepting the target text data from at least one of the header or the tail of the initial text data; A segmentation module, used for performing sentence segmentation processing on the target text data to obtain a plurality of sentences, and determining first sentence feature data corresponding to each of the sentences; A recognition module, used for inputting the first sentence feature data corresponding to each of the sentences into a timeliness recognition model to obtain a first timeliness recognition result corresponding to each of the sentences; The first time validity recognition result is used to represent the validity period of the sentence; A processing module, configured to determine a second timeliness identification result corresponding to the information according to the first timeliness identification result having a shorter validity period; the second timeliness identification result is used to represent the validity period of the information; The aging identification device is also used for: The target text data is matched according to a pre-established corpus matching library to obtain a matching result; the corpus matching library includes a plurality of time-sensitive key sentences; the matching result is obtained by matching the target text data according to the pre-established corpus matching library, including: determining a second similarity between the sentence and the time-sensitive key sentence, and determining the matching result according to the time-sensitive key sentence corresponding to the second similarity greater than a second threshold; Determine a third timeliness identification result of the information according to the matching result and the current time; the third timeliness identification result is used to characterize the validity period of the information; The third timeliness identification result and the second timeliness identification result having the shorter effective time is determined as the updated second timeliness identification result.

11. A computer device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 9.

12. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to implement the method according to any one of claims 1 to 9 when executed by the processor.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.