News processing method and related device based on artificial intelligence

Through scene recognition, key statement extraction and information alignment, the problem of insufficient professional knowledge in news generation in unknown fields is solved, efficient and accurate transmission of news information is achieved, and user experience is improved.

CN118761388BActive Publication Date: 2024-11-22ZHUHAI MEDIA SUNAC TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411219517.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-02
Publication Date
2024-11-22
Estimated Expiration
2044-09-02

AI Technical Summary

Technical Problem

When news generation in unknown fields, insufficient field expertise makes it difficult to accurately understand and extract key information, affecting the quality and accuracy of news generation and failing to effectively pass it to target users.

Method used

By obtaining collection commands, scene recognition and naming, identifying key statements, performing information alignment, determining information transformation relationships, selecting corresponding text generation models, generating target news information, and ensuring the logical and semantic consistency of the information.

Benefits of technology

It improves the consistency and coherence of news information, optimizes the quality of news generation under unknown types, ensures that target users obtain accurate and valuable news content, and improves reading experience and information acquisition efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761388B_ABST
    Figure CN118761388B_ABST
Patent Text Reader

Abstract

The embodiment of the present invention provides a news processing method based on artificial intelligence and a related device, which belongs to the field of artificial intelligence technology. The method includes: performing scene recognition on target news data to obtain a target scene type; when the target scene type is an unknown type, performing scene naming on the target news data to obtain a first scene type, and obtaining a second scene type similar to the first scene type from the known scene types; performing key sentence recognition on the target news data to obtain a first scene sentence and obtain a second scene sentence corresponding to the second scene type; performing information alignment on the first scene sentence and the second scene sentence to obtain distribution alignment information; determining an information conversion relationship according to the distribution alignment information; obtaining a text generation model corresponding to the second scene type, and determining the target news information according to the text generation model and the information conversion relationship; recommending the target news information to the target user, so that the target user obtains relevant information of the target news data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based news processing method and related devices. Background Art

[0002] Automated news writing has significantly improved the efficiency of news production, especially in data-intensive news reporting, such as stock market trends, sports results, and election data. Machines are able to quickly process large amounts of data and generate drafts, giving professional journalists and editors more time for in-depth reporting and content editing. As automated writing technology develops, it can also be applied to generate news reports on less noticeable niche markets, thereby expanding the coverage of news. This includes local news in small cities, industry-specific news, and special reports. However, when applied to news generation in unknown fields, there may be a problem of insufficient domain expertise, which makes it difficult to accurately understand and extract key information, affecting the quality and accuracy of news generation, and ultimately may not be able to effectively deliver the generated news to the target users. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide an artificial intelligence-based news processing method and related devices, aiming to solve the problem in the related technology that when generating news in unknown fields, there may be insufficient domain expertise, which makes it difficult to accurately understand and extract key information, affecting the quality and accuracy of news generation, and thus making it impossible to effectively deliver the generated news to target users.

[0004] In a first aspect, an embodiment of the present invention provides a news processing method based on artificial intelligence, comprising:

[0005] Obtaining a collection command, and obtaining corresponding target news data according to the collection command;

[0006] Performing scene recognition on the target news data to obtain a target scene type corresponding to the target news data;

[0007] When the target scene type is an unknown type, the target news data is scene-named to obtain a first scene type corresponding to the target news data, and a second scene type similar to the first scene type is obtained from known scene types;

[0008] Performing key sentence recognition on the target news data to obtain a first scene sentence corresponding to the first scene type, and obtaining a second scene sentence corresponding to the second scene type;

[0009] Aligning the first scenario statement with the second scenario statement to obtain distribution alignment information between the first scenario statement and the second scenario statement;

[0010] Determine an information conversion relationship between the second scene type and the first scene type according to the distribution alignment information;

[0011] Obtaining a text generation model corresponding to the second scene type, and determining target news information corresponding to the target news data according to the text generation model and the information conversion relationship;

[0012] The target news information is recommended to a target user so that the target user obtains relevant information of the target news data.

[0013] In a second aspect, an embodiment of the present invention provides a news processing device based on artificial intelligence, comprising:

[0014] A data acquisition module, used to obtain an acquisition command and obtain corresponding target news data according to the acquisition command;

[0015] A scene recognition module, used to perform scene recognition on the target news data to obtain a target scene type corresponding to the target news data;

[0016] A type acquisition module, for, when the target scene type is an unknown type, performing scene naming on the target news data, obtaining a first scene type corresponding to the target news data, and obtaining a second scene type similar to the first scene type from known scene types;

[0017] A sentence acquisition module, used to identify key sentences of the target news data, obtain a first scenario sentence corresponding to the first scenario type, and obtain a second scenario sentence corresponding to the second scenario type;

[0018] A data alignment module, used to align information between the first scenario statement and the second scenario statement to obtain distribution alignment information between the first scenario statement and the second scenario statement;

[0019] a relationship determination module, configured to determine an information conversion relationship between the second scene type and the first scene type according to the distribution alignment information;

[0020] A news generation module, used for obtaining a text generation model corresponding to the second scene type, and determining target news information corresponding to the target news data according to the text generation model and the information conversion relationship;

[0021] The news recommendation module is used to recommend the target news information to the target user so that the target user can obtain the relevant information of the target news data.

[0022] In a third aspect, an embodiment of the present invention further provides a terminal device, comprising a processor, a memory, a computer program stored in the memory and executable by the processor, and a data bus for realizing connection and communication between the processor and the memory, wherein when the computer program is executed by the processor, the steps of any one of the artificial intelligence-based news processing methods provided in the specification of the present invention are implemented.

[0023] In a fourth aspect, an embodiment of the present invention further provides a storage medium for computer-readable storage, characterized in that the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the artificial intelligence-based news processing methods provided in the specification of the present invention.

[0024] The embodiment of the present invention provides a news processing method based on artificial intelligence and a related device. The method includes: obtaining an acquisition command, and accurately obtaining corresponding target news data based on the acquisition command, and then performing scene recognition on the target news data to determine the target scene type to which the target news data belongs. When the target scene type is an unknown type, the scene is named for this type of target news data, and a new scene type label is given to it, called the first scene type. At the same time, in order to more effectively process these unknown scenes, a second scene type that is closest to the first scene type is found from a known scene type library, and after the scene type is determined, key sentence recognition is further performed on the target news data to extract key sentences related to the first scene type, namely, first scene sentences, and second scene sentences related to the second scene type are also extracted. In order to more deeply understand the relationship between the two scene types, the first scene sentence and the second scene sentence are information aligned. This alignment process can reveal the distribution alignment information between the two scene sentences, thereby accurately capturing the internal connection and conversion rules between different scenes. The distribution alignment information is of great value in the text generation process. It can not only more accurately reflect the characteristics and requirements of different information scenes, but also better understand and apply the context and language style of different scenes and sentences. Based on the distribution alignment information, the information conversion relationship between the second scene type and the first scene type is determined, and after the information conversion relationship is determined, the text generation model corresponding to the second scene type is selected. Combining the text generation model and the information conversion relationship, the target news information matching the target news data can be accurately generated, thereby ensuring that the generated target news information can be effectively converted and connected based on the known scene type at both the logical and semantic levels. This method not only improves the consistency and coherence of the target news information, but also optimizes the quality of the target news information generated under the unknown type on the basis of the known scene type. Finally, the obtained target news information is recommended to the target user, thereby ensuring that the target user can accurately and efficiently obtain the required news data related information, thereby providing the target user with more accurate and valuable news content, thereby improving their reading experience and information acquisition efficiency. This method also solves the problem in the related art that when generating news in an unknown field, there may be insufficient domain expertise, which makes it difficult to accurately understand and extract key information, affecting the quality and accuracy of news generation, and thus failing to effectively deliver the generated news to the target user. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 A flowchart of a news processing method based on artificial intelligence provided by an embodiment of the present invention;

[0027] Figure 2 A schematic diagram of the module structure of a news processing device based on artificial intelligence provided by an embodiment of the present invention;

[0028] Figure 3 A schematic block diagram of the structure of a terminal device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0031] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0032] The embodiment of the present invention provides a news processing method based on artificial intelligence and a related device. The news processing method based on artificial intelligence can be applied to a terminal device, which can be an electronic device such as a tablet computer, a laptop computer, a desktop computer, a personal digital assistant, and a wearable device. The terminal device can be a server or a server cluster.

[0033] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0034] Please refer to Figure 1 , Figure 1 A flowchart of a news processing method based on artificial intelligence provided by an embodiment of the present invention.

[0035] like Figure 1As shown, the news processing method based on artificial intelligence includes steps S101 to S108.

[0036] Step S101: obtain a collection command, and obtain corresponding target news data according to the collection command.

[0037] Exemplarily, a system or service is set up to receive and process collection commands. This can be an API interface, a message queue, or a task scheduling system. Ensure that the system can accept collection commands from users or other systems, so as to parse the received collection commands and extract key information from the collection commands. For example, the collection command may contain information such as the source, time range, keywords, and categories of the news. After parsing, the above information in the collection command is converted into a specific collection task, so that the information is captured using web crawler technology according to the collection task to obtain the target news data.

[0038] Step S102: performing scene recognition on the target news data to obtain a target scene type corresponding to the target news data.

[0039] Exemplarily, the collected target news data is preprocessed. This includes text cleaning (removing HTML tags, special characters, etc.), word segmentation (splitting text into words or phrases), part-of-speech tagging (marking the part of speech of each word), and removing stop words (removing common words without actual semantics), etc., and then extracting features from the preprocessed target news data using word embedding for scene recognition, and then using the scene recognition model to perform scene recognition on the extracted features, thereby obtaining the target scene type corresponding to the target news data. Among them, the scene recognition model can be a model based on a naive Bayes classifier, a model based on a support vector machine, or a model based on deep learning, etc.

[0040] In some embodiments, the performing scene recognition on the target news data to obtain the target scene type corresponding to the target news data includes: performing word segmentation processing on the target news data to obtain a first keyword, and removing stop words from the first keyword to obtain a second keyword; obtaining a first text vector corresponding to the target news data; obtaining relevant keywords corresponding to each initial scene type in the known scene types and a second text vector corresponding to the initial scene type; calculating a first similarity between the relevant keywords and the second keywords, and calculating a second similarity between the first text representation vector and the second text representation vector; and determining the target scene type corresponding to the target news data from the known scene types based on the first similarity and the second similarity.

[0041] Exemplarily, a target news data text is split into words or phrases using a word segmentation tool. For example, for Chinese text, the jieba word segmentation library can be used; for English text, nltk or spaCy can be used, and then the first keywords corresponding to the target news data are obtained, and a stop word library is used to filter out meaningless words (such as "de", "le") to obtain the second keywords.

[0042] Exemplarily, a first text vector corresponding to the target news data is obtained using a word embedding model (such as Word2Vec, GloVe). And the relevant keywords corresponding to each initial scene type in the known scene types and the second text vectors corresponding to the iconic statements in the initial scene types are obtained.

[0043] Exemplarily, the number of identical keywords between the relevant keywords and the second keywords is obtained to determine the first similarity. And a metric method such as cosine similarity is used to calculate the second similarity between the first text vector and the second text vectors of each initial scene type. Finally, the first similarity and the second similarity are added to obtain the target similarity, and then the maximum value of the target similarity is obtained, and the maximum value of the target similarity is compared with a preset value. When the maximum value of the target similarity is greater than or equal to the preset value, the initial scene type corresponding to the maximum value of the target similarity is determined as the target scene type. When the maximum value of the target similarity is less than the preset value, the target scene type is determined as an unknown type.

[0044] Specifically, calculating the first similarity between the relevant keywords and the second keywords can help identify which core words in the news text are related to a specific scene. Calculating the second similarity between the two text vectors can intuitively reflect the similarity degree between the target news data and the known scene types, so as to more accurately determine the target scene type of the news data. By comprehensively considering the first similarity and the second similarity, the target scene type corresponding to the news data can be judged more comprehensively and accurately. This method can reduce the situation of misclassification due to a relatively high similarity of a single feature and improve the accuracy and stability of scene recognition.

[0045] Step S103, when the target scene type is an unknown type, perform scene naming on the target news data to obtain the first scene type corresponding to the target news data, and obtain the second scene type similar to the first scene type from the known scene types.

[0046] Exemplarily, when the target scene type is any of the known scene types, a text generation model corresponding to the known scene type can be obtained, thereby obtaining the target news information corresponding to the target news data according to the text generation model. However, when the target scene type is an unknown type, it is impossible to directly and accurately obtain the target news information corresponding to the target news data according to the existing text generation model.

[0047] Exemplarily, since the target scene type is an unknown type, it can be determined that the target news data is a newly occurring event, and event extraction can be performed on the target news data, so that the extracted event is determined as the first scene type corresponding to the target news data.

[0048] Exemplarily, in order to ensure that target news information can be subsequently generated based on information in known scene types, a second scene type that is closest to the first scene type is obtained from the known scene types.

[0049] For example, when the maximum value of the target similarity is less than a preset value, the target scene type is determined to be an unknown type, and the initial scene type corresponding to the maximum value of the target similarity under this condition is determined to be the second scene type.

[0050] In some embodiments, obtaining a second scene type similar to the first scene type from known scene types includes: obtaining a third keyword arbitrarily from the second keywords, and obtaining a first target importance corresponding to the third keyword based on the target news data; determining a second target importance corresponding to the related keywords from a database; calculating the same keywords and different keywords between the second keyword and the related keywords; determining a first numerical value corresponding to the same keyword under the second keyword based on the first target importance, and determining a second numerical value corresponding to the different keyword under the second keyword based on the first target importance; determining a third numerical value corresponding to the same keyword under the related keywords based on the second target importance, and determining a fourth numerical value corresponding to the different keyword under the related keywords based on the second target importance; determining the second scene type similar to the first scene type based on the first numerical value, the second numerical value, the third numerical value and the fourth numerical value.

[0051] Exemplarily, a keyword is arbitrarily selected from the second keywords as the third keyword, the TF-IDF value of the third keyword in the target news data is calculated, and then the TF-IDF value is determined as the first target importance corresponding to the third keyword, and the second target importance corresponding to the relevant keyword is obtained from the database.

[0052] Exemplarily, the same keyword corresponding to the second keyword and the related keyword are obtained when they are the same, and the different keywords corresponding to the second keyword and the related keyword are obtained when they are different, and then the sum of the first target importance of the same keywords under the second keyword is calculated to determine the first value, and the sum of the second target importance of the same keywords under the related keywords is calculated to determine the third value.

[0053] Exemplarily, the sum of the first target importances of different keywords under the second keyword is calculated to determine the second value, and the sum of the second target importances of different keywords under the related keyword is calculated to determine the fourth value.

[0054] Exemplarily, a first absolute value of the difference between the first value and the third value, and a second absolute value of the difference between the second value and the fourth value are calculated, and then the initial scene type corresponding to the minimum value of the sum of the first absolute value of the difference and the second absolute value of the difference is determined as a second scene type similar to the first scene type.

[0055] Specifically, by distinguishing the impact of the same keywords and different keywords, we can deeply understand the actual role of keywords in news data. This detailed analysis helps to better identify and understand the context of news data. Detailed numerical calculations help to more accurately match similar scene types and improve the accuracy of data analysis.

[0056] In some embodiments, obtaining the first target importance corresponding to the third keyword based on the target news data includes: obtaining the first connected keyword corresponding to the left connection of the third keyword from the second keyword and obtaining the second connected keyword corresponding to the right connection of the third keyword from the second keyword; calculating the first entropy value corresponding to the third keyword under the first connected keyword and calculating the second entropy value corresponding to the third keyword under the second connected keyword; determining relevant change information corresponding to the third keyword based on the first entropy value and the second entropy value; obtaining important weight information corresponding to the third keyword based on the target news data; and determining the first target importance corresponding to the third keyword based on the relevant change information and the important weight information.

[0057] Exemplarily, a third keyword is obtained randomly from the second keyword, and then the first connection keyword connected to the left of the third keyword and the second connection keyword connected to the right of the third keyword are obtained from the target news data. The first connection keyword is the word located to the left of the third keyword in the target news data, and the second connection keyword is the word located to the right of the third keyword in the target news data.

[0058] Exemplarily, the entropy value is used to measure the uncertainty or amount of information. For each connection keyword, the entropy value of the third keyword under the first connection keyword and the second connection keyword is calculated to indicate the degree of change of the third keyword in the target news data.

[0059] Exemplarily, the probability of the first connection keyword appearing on the left side of the third keyword is obtained, and then the first entropy value is calculated based on the probability. In addition, in order to ensure that the first entropy value does not become zero, a correction term can be added. The correction term is the inverse of the frequency of the first connection keyword appearing on the left side of the third keyword, so that the first entropy value and the correction term are summed to obtain the corrected first entropy value.

[0060] Similarly, the probability that the second connected keyword appears on the right side of the third keyword is obtained, and then the second entropy value is calculated based on the probability. In addition, in order to ensure that the second entropy value does not become zero, a correction term can be added. The correction term is the inverse of the frequency of the second connected keyword appearing on the right side of the third keyword, so that the second entropy value and the correction term are summed to obtain the corrected second entropy value.

[0061] Exemplarily, the first entropy value and the second entropy value are summed to obtain relevant change information corresponding to the third keyword, and the relevant change information is used to characterize the amount of information contained in the target news data by the third keyword. The larger the relevant change information, the greater the amount of information contained in the third keyword, and the smaller the relevant change information, the smaller the amount of information contained in the third keyword.

[0062] Exemplarily, the TF-IDF corresponding to the third keyword is calculated based on the target news data, and the corresponding TF-IDF value is determined as the important weight information, and then the relevant change information and the important weight information are summed, so that the corresponding sum value is determined as the first target importance corresponding to the third keyword.

[0063] Specifically, by obtaining the left and right connected keywords of the third keyword, we can have a more comprehensive understanding of the information content of the third keyword in the target news data, which helps to better understand the meaning and influence of the third keyword in different contexts, and then reveal the changing pattern of the third keyword in different contexts, so as to obtain a more accurate importance of the first target, and then provide good support for the subsequent acquisition of accurate second scene types.

[0064] It should be noted that the above steps can also be used to obtain the second target importance, which will not be repeated here.

[0065] Step S104: perform key sentence recognition on the target news data to obtain a first scenario sentence corresponding to the first scenario type, and obtain a second scenario sentence corresponding to the second scenario type.

[0066] Exemplarily, the most reposted or read news in the target news data is determined as the first scenario sentence corresponding to the first scenario type, and the most reposted or read news in the second scenario type is determined as the second scenario sentence corresponding to the second scenario type.

[0067] Step S105: align information between the first scenario statement and the second scenario statement to obtain distribution alignment information between the first scenario statement and the second scenario statement.

[0068] For example, different scene types have some common features, but each scene type also has its own inherent information. In the scene alignment process, the same information of different scene types is used to link different scene types, thereby effectively reducing the differences between different scene types. Thus, the first scene statement and the second scene statement are subjected to information alignment processing to obtain distribution alignment information between the first scene statement and the second scene statement.

[0069] In some embodiments, the information alignment of the first scenario sentence and the second scenario sentence to obtain the distribution alignment information between the first scenario sentence and the second scenario sentence includes: obtaining the same feature corresponding to the second scenario sentence and the first scenario sentence, and obtaining a first matrix corresponding to the same feature of the second scenario sentence; obtaining a first inherent feature contained only in the second scenario sentence, and obtaining a second matrix corresponding to the first inherent feature; splicing the first matrix and the second matrix to obtain a first text representation corresponding to the second scenario sentence; obtaining a third matrix corresponding to the same feature of the first scenario sentence; obtaining a second inherent feature contained only in the first scenario sentence, and obtaining a fourth matrix corresponding to the second inherent feature; splicing the third matrix and the fourth matrix to obtain the first scene sentence. a second text representation corresponding to a scene sentence; using the first text representation and the second text representation to perform cross-feature filling processing to obtain a first adaptation matrix corresponding to the first scene type acting on the second scene type and a second adaptation matrix corresponding to the second scene type acting on the first scene type; splicing the first adaptation matrix and the first text representation to obtain a third text representation and splicing the second adaptation matrix and the second text representation to obtain a fourth text representation; spatially mapping the third text representation and the fourth text representation to calculate the distribution information between the first scene type and the second scene type; obtaining an alignment matrix corresponding to the second scene sentence and the first scene sentence by adjusting the distribution information; determining the distribution alignment information between the first scene sentence and the second scene sentence according to the alignment matrix.

[0070] Exemplarily, the same features between the second scene sentence and the first scene sentence are identified and extracted, the same features including but not limited to semantic features, keywords, entities, etc., and then the corresponding first matrix in the second scene sentence is determined based on the same features, and the first matrix represents the feature values ​​of the second scene and the wine under the same features.

[0071] Exemplarily, a first inherent feature that only exists in the second scenario sentence is identified and extracted, and a corresponding second matrix is ​​generated for the first inherent feature, and then the first matrix and the second matrix are concatenated to obtain a first text representation of the second scenario sentence.

[0072] Exemplarily, a third matrix corresponding to the first scenario sentence under the same feature is obtained, and the third matrix is ​​used to represent the eigenvalues ​​of the first scenario sentence under the same feature.

[0073] Exemplarily, a second inherent feature that only exists in the first scenario sentence is identified and extracted, and a corresponding fourth matrix is ​​generated for the second inherent feature, and then the third matrix and the fourth matrix are concatenated to obtain a second text representation of the first scenario sentence.

[0074] Exemplarily, a first feature mapping function is determined to map the same features in the first scene type and the second scene type to the inherent features of the second scene type. Then, the obtained feature mapping result is applied to the second text representation to perform feature filling. Finally, through this filling process, a first adaptation matrix of the first scene type to the second scene type is obtained.

[0075] Exemplarily, a second feature mapping function is determined to map the same features in the first scene type and the second scene type to the inherent features of the first scene type. Then, the obtained feature mapping result is applied to the first text representation to perform feature filling. Finally, through this filling process, a second adaptation matrix of the second scene type to the first scene type is obtained.

[0076] Exemplarily, the first adaptation matrix and the first text representation are concatenated to obtain the third text representation, and the second adaptation matrix and the second text representation are concatenated to obtain the fourth text representation.

[0077] Exemplarily, a shared embedding model is used to map the third text representation and the fourth text representation to the same feature space, thereby obtaining first feature information of the third text representation in the feature space and second feature information of the fourth text representation in the feature space, and then calculating the distribution information between the first scene type and the second scene type using Kullback-Leibler divergence by calculating the first feature information and the second feature information, so as to understand the relative position and relationship between the first scene type and the second scene type based on the distribution information, and use it to evaluate the difference between the first scene type and the second scene type.

[0078] Exemplarily, the distribution information can also be represented by a distance matrix or a similarity matrix after spatial mapping, which represents the distribution relationship between the first scene type and the second scene type, and then a method is selected to adjust the distribution information so that the obtained alignment matrix can reflect the optimal alignment effect between the first scene statement and the second scene statement.

[0079] For example, an objective function is set to measure the alignment error, which is usually the difference between the distribution information and the expected alignment. For example, the objective function can be the square error of the distribution information, and then gradient descent, Newton's method or other optimization algorithms are used to minimize the objective function and adjust the alignment matrix. Thus, the corresponding alignment matrix is ​​obtained by continuously adjusting the distribution information, and the alignment matrix represents the alignment relationship between the first scene sentence and the second scene sentence.

[0080] Exemplarily, the purpose of the alignment matrix is ​​to achieve optimal alignment by pairing the sentences in the two scenes to minimize the difference or distance between them. According to the alignment matrix, determine each first scene sentence and its best matching second scene sentence. This can be achieved by analyzing the alignment matrix and finding the column with the highest alignment strength for each first scene sentence, that is, the best matching second scene sentence. Then extract the alignment strength in the alignment matrix and calculate the alignment value of each pair of matching sentences. These alignment values ​​reflect the degree of matching of each sentence pair. Thus, the overall alignment distribution information in the alignment matrix is ​​counted and analyzed, such as calculating the mean, variance, etc. of the alignment strength. These statistical information helps to understand the overall alignment between the first scene sentence and the second scene sentence, and then output the extracted alignment information and distribution results in an appropriate format. For example, generate an alignment pair list, etc. to facilitate further analysis and interpretation of the distribution alignment information. Thus, the distribution alignment information between the first scene sentence and the second scene sentence is obtained, and their matching and overall alignment distribution can be understood.

[0081] Specifically, the same features and inherent features between the first scene sentence and the second scene sentence are identified and processed to improve the accuracy of text representation, and then through splicing and cross-feature filling, a text representation containing more information is generated to improve the ability to understand the scene. Thus, through spatial mapping and adjustment of distribution information, a more accurate alignment matrix and distribution alignment information are obtained to help deeply analyze the relationship between scenes, which in turn helps to improve the accuracy of sentence alignment and enhance the ability to understand the relationship between scenes.

[0082] Step S106: Determine an information conversion relationship between the second scene type and the first scene type according to the distribution alignment information.

[0083] Exemplarily, the distribution alignment information is used to describe the pairing relationship and alignment strength between the first scene statement and the second scene statement, and then by comparing the distribution alignment information of the first scene type and the second scene type, the feature mapping relationship of each scene type is determined. If the distribution alignment information shows that the relationship between the scene types is a linear relationship, linear regression or other linear models can be used to construct the mapping. If the distribution alignment information shows a nonlinear relationship, a neural network or other nonlinear model can be used to capture complex mapping relationships.

[0084] Exemplarily, representative sample pairs of the first scene type and the second scene type are selected from the distribution alignment information. These samples correspond to sentence pairs with high alignment strength to ensure the accuracy of the conversion relationship. The selected sample pairs are then used to train the conversion model. The model can be a linear transformation matrix or other suitable conversion function, so that a part of the data in the sample pair is used to perform model verification to ensure that the conversion model accurately reflects the information conversion relationship between the first scene type and the second scene type.

[0085] In some embodiments, determining the information conversion relationship between the second scene type and the first scene type according to the distribution alignment information includes: using the distribution alignment information to determine the first scene representation corresponding to the first scene sentence in the first scene type, and using the distribution alignment information to determine the second scene representation corresponding to the second scene sentence in the second scene type; obtaining a third scene sentence from a database, and calculating a third similarity between the third scene sentence and the first scene sentence according to the first scene representation; obtaining a fourth scene sentence from a database, and calculating a fourth similarity between the fourth scene sentence and the second scene sentence according to the second scene representation; obtaining a first associated sentence corresponding to the first scene sentence according to the third similarity, and obtaining a second associated sentence corresponding to the second scene sentence according to the fourth similarity; obtaining a first summarized text corresponding to the first associated sentence, and obtaining a second summarized text corresponding to the second associated sentence; obtaining a first conversion relationship between the first associated sentence and the first summarized text through model training, and obtaining a second conversion relationship between the second associated sentence and the second summarized text through model training; and obtaining the information conversion relationship between the second scene type and the first scene type by combining the first conversion relationship and the second conversion relationship.

[0086] Exemplarily, alignment data related to the first scenario type is extracted from the distribution alignment information. This includes the alignment strength between each first scenario statement and the second scenario statement. Based on the alignment strength, the first scenario representation of each first scenario statement is calculated. Typically, this can be achieved by weighted averaging or other aggregation methods using the information in the alignment matrix. Similarly, alignment data related to the second scenario type is extracted from the distribution alignment information. This includes the alignment strength between each second scenario statement and the first scenario statement, and then based on the alignment strength, the second scenario representation of each second scenario statement is calculated. Typically, this can also be achieved by weighted averaging or other aggregation methods using the information in the alignment matrix.

[0087] Exemplarily, a third scene sentence related to the first scene type is retrieved from the database, and a third similarity between the features of the first scene representation and the third scene sentence is calculated using a metric such as cosine similarity or Euclidean distance.

[0088] Exemplarily, a fourth scene sentence related to the second scene type is retrieved from the database, and a fourth similarity between the features of the second scene representation and the fourth scene sentence is calculated using a metric such as cosine similarity or Euclidean distance.

[0089] Exemplarily, according to the third similarity, the third scenario sentence having the highest similarity to the first scenario sentence is selected as the first associated sentence corresponding to the first scenario sentence. Similarly, according to the fourth similarity, the fourth scenario sentence having the highest similarity to the second scenario sentence is selected as the second associated sentence corresponding to the second scenario sentence.

[0090] Exemplarily, a first summary text corresponding to the first associated sentence is retrieved from a database or related resources, and the first summary text may be a summary text related to the first associated sentence. A second summary text of the second associated sentence is retrieved from a database or related resources, and the second summary text is a summary text related to the second associated sentence.

[0091] Exemplarily, the first associated sentence and the first summarized text are used as training data to train the model. The model may be a regression model, a neural network, etc., for learning the first conversion relationship between the first associated sentence and the first summarized text.

[0092] Exemplarily, the second associated sentence and the second summarized text are used as training data to train the model. The model can be a regression model, a neural network, etc., for learning the second conversion relationship between the second associated sentence and the second summarized text.

[0093] Exemplarily, the difference information between the first conversion relationship and the second conversion relationship is calculated, and then the difference information is determined as the information conversion relationship between the second scene type and the first scene type, so that after obtaining the information conversion relationship, the text generation result corresponding to the first scene type can be obtained by using the text generation model corresponding to the second scene type.

[0094] For example, the second conversion relationship is expressed as the sum or product of the first conversion relationship and the unknown relationship, and then the unknown relationship is determined as the information conversion relationship between the second scene type and the first scene type.

[0095] Exemplarily, by using distribution alignment information to determine the scene representation, information in different scene types can be captured and aligned more accurately. This helps to improve the understanding of scene sentences, so that the meaning of sentences can be better identified and matched in the comparison of multiple scenes. Then, through the similarity calculation based on the scene representation (the third similarity and the fourth similarity), the most relevant sentences can be effectively found in a large-scale database. This makes the similarity evaluation more detailed and targeted, so that relevant related sentences can be obtained more accurately. Thus, the conversion relationship obtained by training the model makes the information conversion from the first scene type to the second scene type controllable and repeatable. This training can ensure that the model provides consistent and accurate information conversion in practical applications. Thereby providing good support for the subsequent acquisition of target news information.

[0096] In some embodiments, the calculating the third similarity between the third scenario sentence and the first scenario sentence according to the first scenario representation includes: obtaining a third scenario representation corresponding to the third scenario sentence, and performing similarity calculation according to the first scenario representation and the third scenario representation to obtain a first information similarity; obtaining a first word number of the same words between the third scenario sentence and the first scenario sentence, and determining a second information similarity according to the first word data and a second word number corresponding to the first scenario words; obtaining a third word number of the same words between the third scenario sentence and the second scenario sentence, and determining a third information similarity according to the third word number and a fourth word number corresponding to the second scenario words; obtaining a fifth word number corresponding to the third scenario sentence and a sixth word number corresponding to the second scenario sentence, and determining a fourth information similarity according to the fifth word number and the sixth word number; determining the third similarity according to the first information similarity, the second information similarity, the third information similarity, and the fourth information similarity; wherein the third similarity is obtained according to the following formula:

[0097] ;

[0098] in, represents the third similarity, represents a first weight corresponding to the first information similarity, represents a second weight corresponding to the second information similarity, represents a third weight corresponding to the third information similarity, represents a fourth weight corresponding to the fourth information similarity, The first scenario indicates that, The third scenario indicates that, represents the number of the first word, represents the number of the second word, represents the number of the third word, represents the number of the fourth word, represents the number of the fifth word, Indicates the number of the sixth word.

[0099] Exemplarily, necessary preprocessing is performed on the third scene sentence, such as word segmentation, removal of stop words, etc. The bag-of-words model, TF-IDF, word2vec or other advanced text representation methods (such as BERT embedding) are used to extract the features of the third scene sentence to form a third scene representation.

[0100] Exemplarily, cosine similarity, Euclidean distance or other similarity calculation methods are used to compare the similarities between the first scene representation and the third scene representation to obtain the first information similarity, and then the third scene sentence and the first scene sentence are segmented, and the number of identical words in the two is counted to obtain the first word number, and then the second information similarity is determined by calculating the ratio of the first word number to the second word number, or other similarity calculation methods based on word overlap.

[0101] Exemplarily, similarly, the third scene sentence and the second scene sentence are segmented, and the number of identical words in the two is counted to obtain the third word number. The number of all words in the second scene sentence is counted to obtain the fourth word number, and the third information similarity is determined by calculating the ratio of the third word number to the fourth word number or other similarity calculation methods.

[0102] Exemplarily, the number of words in the third scene sentence and the second scene sentence are counted respectively to obtain the fifth number of words and the sixth number of words, and then the fourth information similarity can be calculated by comparing the difference or ratio between the fifth number of words and the sixth number of words.

[0103] For example, the four information similarities can be combined by weighted average, linear combination or other fusion methods to obtain the final third similarity. The weight can be adjusted according to the actual application scenario and requirements. Among them, the third similarity can be obtained according to the following formula:

[0104] ;

[0105] in, represents the third similarity, represents the first weight corresponding to the first information similarity, represents the second weight corresponding to the second information similarity, represents the third weight corresponding to the third information similarity, represents the fourth weight corresponding to the fourth information similarity, Indicates the first scene representation, The third scene shows that, Indicates the number of the first word, Indicates the number of the second word, Indicates the number of the third word, Indicates the number of the fourth word, Indicates the number of the fifth word, Indicates the number of the sixth word.

[0106] Specifically, by comparing the similarity between the first scene representation and the third scene representation, their similarity in the semantic space can be evaluated. This helps to identify scenes with similar content. The similarity calculation can quantify the proximity of the two scenes in the feature space and improve the accuracy of scene matching. Calculating the number of identical words can reflect the degree of vocabulary overlap between the two, thereby evaluating the similarity of the text content. By comparing the word overlap between the third scene sentence and the second scene sentence, we can understand the vocabulary sharing in different scenes, further refine the similarity evaluation, help identify the symmetry or consistency of vocabulary in different scenes, and provide more detailed information about the similarity between scenes. Counting the number of words can help analyze the content coverage and vocabulary richness of the text, thereby providing a quantitative evaluation of text similarity. And by comprehensively considering multiple similarity indicators, a more comprehensive and accurate similarity evaluation can be provided.

[0107] Step S107: obtaining a text generation model corresponding to the second scene type, and determining target news information corresponding to the target news data according to the text generation model and the information conversion relationship.

[0108] Exemplarily, a corresponding text generation model is obtained from an existing model library based on the second scenario type. For example, if the second scenario type is news generation, a modern language model such as GPT-3, GPT-4, or a model specifically for news generation tasks (such as a model based on T5) can be selected.

[0109] Exemplarily, the target news data is input into the selected text generation model to obtain initial news information, and the initial news information is converted according to the information conversion relationship to obtain the target news information corresponding to the target news data.

[0110] Step S108: recommend the target news information to a target user, so that the target user obtains relevant information of the target news data.

[0111] Exemplarily, user data (such as user behavior data, interest preferences, historical browsing records, etc.) is collected and stored to identify the user's interests, preferences and needs, so as to screen out target users corresponding to the target news information based on the user's interests, preferences and needs, and then recommend the target news information to the target users, so that the target users obtain relevant information of the target news data. The recommendation method can deliver the recommended content to the target users through news applications, email subscriptions, social media and other channels.

[0112] The embodiment of the present invention provides a news processing method based on artificial intelligence and a related device. The method includes: obtaining an acquisition command, and accurately obtaining corresponding target news data based on the acquisition command, and then performing scene recognition on the target news data to determine the target scene type to which the target news data belongs. When the target scene type is an unknown type, the scene is named for this type of target news data, and a new scene type label is given to it, called the first scene type. At the same time, in order to more effectively process these unknown scenes, a second scene type that is closest to the first scene type is found from a known scene type library, and after the scene type is determined, key sentence recognition is further performed on the target news data to extract key sentences related to the first scene type, namely, first scene sentences, and second scene sentences related to the second scene type are also extracted. In order to more deeply understand the relationship between the two scene types, the first scene sentence and the second scene sentence are information aligned. This alignment process can reveal the distribution alignment information between the two scene sentences, thereby accurately capturing the internal connection and conversion rules between different scenes. The distribution alignment information is of great value in the text generation process. It can not only more accurately reflect the characteristics and requirements of different information scenes, but also better understand and apply the context and language style of different scenes and sentences. Based on the distribution alignment information, the information conversion relationship between the second scene type and the first scene type is determined, and after the information conversion relationship is determined, the text generation model corresponding to the second scene type is selected. Combining the text generation model and the information conversion relationship, the target news information matching the target news data can be accurately generated, thereby ensuring that the generated target news information can be effectively converted and connected based on the known scene type at both the logical and semantic levels. This method not only improves the consistency and coherence of the target news information, but also optimizes the quality of the target news information generated under the unknown type on the basis of the known scene type. Finally, the obtained target news information is recommended to the target user, thereby ensuring that the target user can accurately and efficiently obtain the required news data related information, thereby providing the target user with more accurate and valuable news content, thereby improving their reading experience and information acquisition efficiency. This method also solves the problem in the related art that when generating news in an unknown field, there may be insufficient domain expertise, which makes it difficult to accurately understand and extract key information, affecting the quality and accuracy of news generation, and thus failing to effectively deliver the generated news to the target user.

[0113] See also Figure 2 , Figure 2An artificial intelligence-based news processing device 200 is provided in an embodiment of the present application. The artificial intelligence-based news processing device 200 includes a data acquisition module 201, a scene recognition module 202, a type acquisition module 203, a sentence acquisition module 204, a data alignment module 205, a relationship determination module 206, a news generation module 207, and a news recommendation module 208, wherein the data acquisition module 201 is used to obtain an acquisition command and obtain corresponding target news data according to the acquisition command; the scene recognition module 202 is used to perform scene recognition on the target news data and obtain a target scene type corresponding to the target news data; the type acquisition module 203 is used to perform scene naming on the target news data when the target scene type is an unknown type, obtain a first scene type corresponding to the target news data, and obtain a second scene type similar to the first scene type from the known scene types; A sentence acquisition module 204 is used to identify key sentences of the target news data, obtain the first scenario sentence corresponding to the first scenario type, and obtain the second scenario sentence corresponding to the second scenario type; a data alignment module 205 is used to align the first scenario sentence with the second scenario sentence, and obtain the distribution alignment information between the first scenario sentence and the second scenario sentence; a relationship determination module 206 is used to determine the information conversion relationship between the second scenario type and the first scenario type according to the distribution alignment information; a news generation module 207 is used to obtain a text generation model corresponding to the second scenario type, and determine the target news information corresponding to the target news data according to the text generation model and the information conversion relationship; a news recommendation module 208 is used to recommend the target news information to a target user, so that the target user obtains the relevant information of the target news data.

[0114] In some embodiments, the artificial intelligence-based news processing apparatus 200 may be applied to a terminal device.

[0115] It should be noted that technicians in the relevant field can clearly understand that for the convenience and brevity of description, the specific working process of the artificial intelligence-based news processing device 200 described above can refer to the corresponding process in the aforementioned artificial intelligence-based news processing method embodiment, and will not be repeated here.

[0116] See also Figure 3 , Figure 3 A schematic block diagram of the structure of a terminal device provided in an embodiment of the present invention.

[0117] like Figure 3As shown, the terminal device 300 includes a processor 301 and a memory 302 , and the processor 301 and the memory 302 are connected via a bus 303 , such as an I2C (Inter-integrated Circuit) bus.

[0118] Specifically, the processor 301 is used to provide computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit (CPU), and the processor 301 can also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0119] Specifically, the memory 302 may be a Flash chip, a read-only memory (ROM) disk, an optical disk, a USB flash drive, or a mobile hard disk.

[0120] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the embodiment of the present invention, and does not constitute a limitation on the terminal device to which the embodiment of the present invention is applied. The specific server may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0121] The processor is used to run the computer program stored in the memory, and implement any one of the artificial intelligence-based news processing methods provided by the embodiments of the present invention when executing the computer program.

[0122] In one embodiment, the processor is used to run a computer program stored in the memory, and implements the following steps when executing the computer program:

[0123] Obtaining a collection command, and obtaining corresponding target news data according to the collection command;

[0124] Performing scene recognition on the target news data to obtain a target scene type corresponding to the target news data;

[0125] When the target scene type is an unknown type, the target news data is scene-named to obtain a first scene type corresponding to the target news data, and a second scene type similar to the first scene type is obtained from known scene types;

[0126] Performing key sentence recognition on the target news data to obtain a first scene sentence corresponding to the first scene type, and obtaining a second scene sentence corresponding to the second scene type;

[0127] Aligning the first scenario statement with the second scenario statement to obtain distribution alignment information between the first scenario statement and the second scenario statement;

[0128] Determine an information conversion relationship between the second scene type and the first scene type according to the distribution alignment information;

[0129] Obtaining a text generation model corresponding to the second scene type, and determining target news information corresponding to the target news data according to the text generation model and the information conversion relationship;

[0130] The target news information is recommended to a target user so that the target user obtains relevant information of the target news data.

[0131] It should be noted that technicians in the relevant field can clearly understand that for the convenience and brevity of description, the specific working process of the terminal device described above can refer to the corresponding process in the aforementioned artificial intelligence-based news processing method embodiment, and will not be repeated here.

[0132] An embodiment of the present invention also provides a storage medium for computer-readable storage, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the artificial intelligence-based news processing methods provided in the description of the embodiment of the present invention.

[0133] The storage medium may be an internal storage unit of the terminal device in the aforementioned embodiment, such as a hard disk or memory of the terminal device. The storage medium may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal device.

[0134] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware embodiment, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transient medium). As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0135] It should be understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0136] The serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A news processing method based on artificial intelligence, characterized in that: The method comprises: Obtaining a collection command, and obtaining corresponding target news data according to the collection command; Performing scene recognition on the target news data to obtain a target scene type corresponding to the target news data; When the target scene type is an unknown type, the target news data is scene-named to obtain a first scene type corresponding to the target news data, and a second scene type similar to the first scene type is obtained from known scene types; Performing key sentence recognition on the target news data to obtain a first scene sentence corresponding to the first scene type, and obtaining a second scene sentence corresponding to the second scene type; Aligning the first scenario statement with the second scenario statement to obtain distribution alignment information between the first scenario statement and the second scenario statement; Determine an information conversion relationship between the second scene type and the first scene type according to the distribution alignment information; Obtaining a text generation model corresponding to the second scene type, and determining target news information corresponding to the target news data according to the text generation model and the information conversion relationship; The target news information is recommended to a target user so that the target user obtains relevant information of the target news data.

2. The method according to claim 1, characterized in that The performing scene recognition on the target news data to obtain the target scene type corresponding to the target news data includes: Performing word segmentation processing on the target news data to obtain a first keyword, and performing stop word removal processing on the first keyword to obtain a second keyword; Obtaining a first text vector corresponding to the target news data; Obtaining relevant keywords corresponding to each initial scene type in the known scene types and a second text vector corresponding to the initial scene type; Calculating a first similarity between the related keyword and the second keyword, and calculating a second similarity between the first text vector and the second text vector; The target scene type corresponding to the target news data is determined from the known scene types according to the first similarity and the second similarity.

3. The method according to claim 2, characterized in that The obtaining a second scene type similar to the first scene type from known scene types includes: Obtaining any third keyword from the second keywords, and obtaining the first target importance corresponding to the third keyword according to the target news data; Determine the second target importance corresponding to the relevant keyword from the database; Calculating the same keywords and different keywords between the second keyword and the related keyword; Determine a first value corresponding to the same keyword under the second keyword according to the first target importance, and determine a second value corresponding to the different keyword under the second keyword according to the first target importance; Determine a third value corresponding to the same keyword under the related keywords according to the second target importance, and determine a fourth value corresponding to the different keyword under the related keywords according to the second target importance; The second scene type that is similar to the first scene type is determined according to the first value, the second value, the third value, and the fourth value.

4. The method according to claim 3, characterized in that The obtaining, according to the target news data, a first target importance corresponding to the third keyword includes: Obtain a first connection keyword corresponding to the left side connection of the third keyword from the second keyword and obtain a second connection keyword corresponding to the right side connection of the third keyword from the second keyword; Calculating a first entropy value corresponding to the third keyword under the first connection keyword and calculating a second entropy value corresponding to the third keyword under the second connection keyword; Determine relevant change information corresponding to the third keyword according to the first entropy value and the second entropy value; Obtaining important weight information corresponding to the third keyword according to the target news data; The first target importance corresponding to the third keyword is determined according to the relevant change information and the importance weight information.

5. The method according to claim 1, characterized in that The step of aligning information between the first scenario statement and the second scenario statement to obtain distribution alignment information between the first scenario statement and the second scenario statement includes: Obtaining the same features corresponding to the second scenario sentence and the first scenario sentence, and obtaining a first matrix corresponding to the same features of the second scenario sentence; Obtaining a first inherent feature contained only in the second scene sentence, and obtaining a second matrix corresponding to the first inherent feature; Concatenating the first matrix and the second matrix to obtain a first text representation corresponding to the second scene sentence; Obtain a third matrix corresponding to the same feature of the first scene sentence; Obtaining a second intrinsic feature contained only in the first scene sentence, and obtaining a fourth matrix corresponding to the second intrinsic feature; Concatenating the third matrix and the fourth matrix to obtain a second text representation corresponding to the first scene sentence; Performing cross-feature filling processing using the first text representation and the second text representation to obtain a first adaptation matrix corresponding to the first scene type acting on the second scene type and a second adaptation matrix corresponding to the second scene type acting on the first scene type; concatenating the first adaptation matrix and the first text representation to obtain a third text representation and concatenating the second adaptation matrix and the second text representation to obtain a fourth text representation; Performing spatial mapping on the third text representation and the fourth text representation, and calculating distribution information between the first scene type and the second scene type; Obtaining an alignment matrix corresponding to the second scenario sentence and the first scenario sentence by adjusting the distribution information; The distribution alignment information between the first scenario statement and the second scenario statement is determined according to the alignment matrix.

6. The method according to claim 1, characterized in that The determining, according to the distribution alignment information, an information conversion relationship between the second scene type and the first scene type includes: Determine a first scene representation corresponding to the first scene statement in the first scene type by using the distribution alignment information, and determine a second scene representation corresponding to the second scene statement in the second scene type by using the distribution alignment information; Obtaining a third scene sentence from the database, and calculating a third similarity between the third scene sentence and the first scene sentence according to the first scene representation; Obtaining a fourth scenario statement from the database, and calculating a fourth similarity between the fourth scenario statement and the second scenario statement according to the second scenario representation; Obtaining a first associated sentence corresponding to the first scenario sentence according to the third similarity, and obtaining a second associated sentence corresponding to the second scenario sentence according to the fourth similarity; Obtaining a first summarized text corresponding to the first associated sentence, and obtaining a second summarized text corresponding to the second associated sentence; Obtaining a first conversion relationship between the first associated sentence and the first summarized text through model training, and obtaining a second conversion relationship between the second associated sentence and the second summarized text through model training; The first conversion relationship and the second conversion relationship are combined to obtain the information conversion relationship between the second scene type and the first scene type.

7. The method according to claim 6, characterized in that The calculating, according to the first scene representation, a third similarity between the third scene statement and the first scene statement comprises: Obtaining a third scene representation corresponding to the third scene sentence, and performing similarity calculation based on the first scene representation and the third scene representation to obtain first information similarity; Obtaining a first word quantity of the third scenario sentence and the first scenario sentence that are identical, and determining a second information similarity according to the first word quantity and a second word quantity corresponding to the first scenario sentence; Obtaining the number of third words that are the same as the second scene sentence in the third scene sentence, and determining the third information similarity according to the number of third words and the number of fourth words corresponding to the second scene sentence; Obtaining a fifth word quantity corresponding to the third scenario sentence and a sixth word quantity corresponding to the second scenario sentence, and determining a fourth information similarity according to the fifth word quantity and the sixth word quantity; Determine the third similarity according to the first information similarity, the second information similarity, the third information similarity, and the fourth information similarity; The third similarity is obtained according to the following formula: ; in, represents the third similarity, represents a first weight corresponding to the first information similarity, represents a second weight corresponding to the second information similarity, represents a third weight corresponding to the third information similarity, represents a fourth weight corresponding to the fourth information similarity, The first scenario indicates that, The third scenario indicates that, represents the number of the first word, represents the number of the second word, represents the number of the third word, represents the number of the fourth word, represents the number of the fifth word, Indicates the number of the sixth word.

8. A news processing device based on artificial intelligence, characterized in that: include: A data acquisition module, used to obtain an acquisition command and obtain corresponding target news data according to the acquisition command; A scene recognition module, used to perform scene recognition on the target news data to obtain a target scene type corresponding to the target news data; A type acquisition module, for, when the target scene type is an unknown type, performing scene naming on the target news data, obtaining a first scene type corresponding to the target news data, and obtaining a second scene type similar to the first scene type from known scene types; A sentence acquisition module, used to identify key sentences of the target news data, obtain a first scenario sentence corresponding to the first scenario type, and obtain a second scenario sentence corresponding to the second scenario type; A data alignment module, used to align information between the first scenario statement and the second scenario statement to obtain distribution alignment information between the first scenario statement and the second scenario statement; a relationship determination module, configured to determine an information conversion relationship between the second scene type and the first scene type according to the distribution alignment information; A news generation module, used for obtaining a text generation model corresponding to the second scene type, and determining target news information corresponding to the target news data according to the text generation model and the information conversion relationship; The news recommendation module is used to recommend the target news information to the target user so that the target user can obtain the relevant information of the target news data.

9. A terminal device, characterized in that: The terminal device includes a processor and a memory; The memory is used to store computer programs; The processor is used to execute the computer program and implement the artificial intelligence-based news processing method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer storage medium for computer storage, characterized in that: The computer storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the artificial intelligence-based news processing method described in any one of claims 1 to 7.