Personalized matching method and system of digital cultural education resources based on AIGC
By applying AIGC technology in the big data push matching system, combining user language habits and conversation text semantics to generate personalized recommendations, the existing system's shortcomings in handling user language habits and deep semantics are solved, and more accurate content push is achieved .
Patent Information
- Application Number
- CN202410897162.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-05
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-07-05
AI Technical Summary
The existing big data push matching system ignores user language habits and conversational state styles when processing digital cultural and educational interaction logs, and semantic mining only focuses on the surface meaning, resulting in the push effect not being accurate and personalized enough.
Using AIGC-based method, by analyzing the digital cultural and educational interaction logs, identifying the conversation state style, performing semantic mining of conversation text, generating user language habit descriptions and linear knowledge representations of conversation interests, combining user push preference knowledge representations to perform knowledge cross-operation, and determining Targeted subscription content push tags.
It improves the accuracy and personalization of push, can better understand user needs, capture deep semantics, and provide more accurate content recommendations.
Smart Images

Figure CN118446856B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of digital culture and education technology, and more specifically, to a personalized matching method and system for digital culture and education resources based on AIGC. Background Art
[0002] With the rapid development of digital technology and network communications, digital cultural and educational resources are becoming increasingly abundant, and users' demand for personalized and precise content is also growing. In order to meet this demand, big data push matching systems are widely used in the field of digital culture and education. By analyzing users' behavioral data, preference settings, etc., they push content that meets their interests and needs to users.
[0003] However, when processing digital cultural and educational interaction logs, existing big data push matching systems often ignore the impact of user language habits and conversational style on push effects. User language habits and conversational style are important aspects that reflect their personality and interests, and are of great significance for accurate push. At the same time, when performing semantic mining, existing systems often only focus on the surface meaning of the text, but ignore the deep semantic connection between texts, which to some extent limits the accuracy and personalization of push. Summary of the invention
[0004] In order to improve the technical problems existing in the relevant technologies, this application provides a personalized matching method and system for digital cultural education resources based on AIGC.
[0005] In a first aspect, the embodiment of the present application provides a personalized matching method for digital cultural education resources based on AIGC, which is applied to a big data push matching system, and the method includes:
[0006] When parsing the digital cultural and educational interaction log, the user language habit description of the corresponding parsing period is obtained according to the conversation state style identified in each parsing period;
[0007] Collecting digital cultural and educational texts from the digital cultural and educational interaction logs, and performing conversation text semantic mining on each collected digital cultural and educational text to obtain conversation text embedded semantics corresponding to each digital cultural and educational text;
[0008] Based on the obtained description of each user's language habits, a corresponding linear knowledge representation of conversation interests is obtained, and based on the obtained embedded semantics of each conversation text, a corresponding knowledge representation of user push preferences is obtained;
[0009] A knowledge cross-operation is performed on the user push preference knowledge representation and the conversation interest linear knowledge representation to obtain corresponding information recommendation auxiliary decision-making knowledge, and based on the information recommendation auxiliary decision-making knowledge, a directional subscription content push tag of the digital cultural and educational interaction log is determined.
[0010] In combination with the first aspect, in a possible implementation of the first aspect, the method further includes: determining whether the number of digital cultural and educational texts collected during the target subscription content push phase reaches a preset number threshold;
[0011] Then, the step of obtaining the corresponding linear knowledge representation of conversation interest based on the obtained language habit descriptions of each user, and obtaining the corresponding knowledge representation of user push preference based on the obtained embedded semantics of each conversation text, includes: when the preset number threshold is reached, obtaining the corresponding linear knowledge representation of conversation interest based on the language habit descriptions of each user obtained in the target subscription content push stage, and obtaining the corresponding knowledge representation of user push preference based on the embedded semantics of each conversation text obtained in the target subscription content push stage;
[0012] Based on the information recommendation auxiliary decision-making knowledge, determining the targeted subscription content push tag of the digital cultural and educational interaction log includes: based on the information recommendation auxiliary decision-making knowledge, determining the targeted subscription content push tag of the digital cultural and educational interaction log in the target subscription content push stage.
[0013] In combination with the first aspect, in a possible implementation of the first aspect, conversation text semantic mining is performed on each collected digital cultural and educational text to obtain conversation text embedded semantics corresponding to each digital cultural and educational text, including:
[0014] For each of the digital cultural and educational texts, the following processing is performed:
[0015] For a digital cultural and educational text, performing text word vector mining on the digital cultural and educational text through a deep residual model node to obtain an original text word vector relationship network of the digital cultural and educational text;
[0016] The original text word vector relationship network is entered into several cascaded vector relationship network processing nodes to obtain the final text word vector relationship network of the last vector relationship network processing node, and the conversation text embedded semantics is obtained based on the final text word vector relationship network; wherein the processing process of each vector relationship network processing node includes upsampling the entered text word vector relationship network to obtain a first transition text word vector relationship network, performing an expansion sliding filtering operation on the first transition text word vector relationship network to obtain a second transition text word vector relationship network, and performing feature focusing-based processing on the second transition text word vector relationship network to obtain a third transition text word vector relationship network, and performing feature optimization on the third transition text word vector relationship network and the entered text word vector relationship network to obtain a final text word vector relationship network.
[0017] In combination with the first aspect, in a possible implementation of the first aspect, for a digital cultural and educational text, performing text word vector mining on the digital cultural and educational text through a deep residual model node to obtain an original text word vector relationship network of the digital cultural and educational text includes:
[0018] Acquire dialogue information corresponding to a user emotion description dimension from dialogue information of a plurality of description dimensions corresponding to the one digital cultural and educational text, wherein the plurality of description dimensions include a user emotion description dimension and a word and sentence logic description dimension;
[0019] Through the deep residual model node, text word vector mining is performed on the dialogue information corresponding to the user emotion description dimension to obtain the original text word vector relationship network.
[0020] In combination with the first aspect, in a possible implementation of the first aspect, based on the obtained embedded semantics of each conversation text, a corresponding user push preference knowledge representation is obtained by any one of the following methods:
[0021] Integrate the embedded semantics of each conversation text to obtain the user push preference knowledge representation;
[0022] Performing average processing on the embedded semantics of each conversation text to obtain the knowledge representation of the user push preference;
[0023] Down-sampling processing is performed on the embedded semantics of each conversation text to obtain the knowledge representation of the user push preference.
[0024] In combination with the first aspect, in a possible implementation of the first aspect, when parsing the digital cultural and educational interaction log, a user language habit description of the corresponding parsing period is obtained according to the session state style identified in each parsing period, including:
[0025] For each parsing cycle, the following operations are performed:
[0026] Determining whether a semantic gap occurs in a parsing cycle based on a difference between a digital cultural and educational text parsed in a parsing cycle and a digital cultural and educational text parsed in at least one past parsing cycle closest to the parsing cycle;
[0027] Based on the judgment point of view, the one parsing cycle is marked to obtain the corresponding user language habit description; wherein, when the judgment point of view is continuous, the first language habit variable is marked, and when the judgment point of view is semantic fault, the second language habit variable is marked.
[0028] In combination with the first aspect, in a possible implementation of the first aspect, obtaining a corresponding conversation interest linear knowledge representation based on the obtained language habit description of each user includes:
[0029] Arrange the obtained language habit descriptions of each user according to the priority between the corresponding analysis cycles to obtain a user language habit description sequence;
[0030] Based on preset description variable mapping features, determining keyword mapping vectors corresponding to a plurality of interest keywords from the user language habit description sequence;
[0031] Based on the keyword mapping vectors of each of the multiple interest keywords, the conversation interest linear knowledge representation is obtained.
[0032] In combination with the first aspect, in a possible implementation of the first aspect, a knowledge cross-operation is performed on the user push preference knowledge representation and the session interest linear knowledge representation to obtain corresponding information recommendation auxiliary decision-making knowledge, and based on the information recommendation auxiliary decision-making knowledge, a targeted subscription content push tag of the digital cultural and educational interaction log is determined, including:
[0033] Performing a knowledge dimension adjustment operation on the linear knowledge representation of the conversation interest so that the knowledge dimension of the linear knowledge representation of the conversation interest is the same as the knowledge dimension of the user push preference knowledge representation;
[0034] Based on the linkage combination result between the user push preference knowledge representation and the session interest linear knowledge representation, the information recommendation auxiliary decision-making knowledge is obtained;
[0035] The information recommendation auxiliary decision knowledge is entered into a push decision tree network to obtain a directional subscription content push tag output by the push decision tree network.
[0036] In combination with the first aspect, in a possible implementation manner of the first aspect, the method is implemented by a big data push matching network, and the debugging method of the big data push matching network includes:
[0037] Through several sample digital cultural and educational interaction logs, the big data push matching network is debugged multiple times in a loop; wherein, in each loop debugging, the following processing is implemented:
[0038] Collect digital cultural and educational texts from the input sample digital cultural and educational interaction logs, and perform conversation text semantic mining on each collected digital cultural and educational text to obtain the corresponding conversation text embedding semantics, and, according to the semantic fault state of each parsing cycle in the sample digital cultural and educational interaction logs, obtain the user language habit description of the corresponding parsing cycle;
[0039] Based on the obtained semantic embedding of each conversation text, the corresponding user push preference knowledge representation is obtained, and based on the obtained description of each user's language habits, the corresponding conversation interest linear knowledge representation is obtained;
[0040] Performing a knowledge cross-operation on the user push preference knowledge representation and the session interest linear knowledge representation to obtain corresponding information recommendation decision-making assistance knowledge, and determining a targeted subscription content push tag based on the information recommendation decision-making assistance knowledge;
[0041] Based on the a priori subscription content push tags and the directional subscription content push tags of the several sample digital cultural and educational interaction logs, a network debugging error is determined, and a neural network weight is improved based on the network debugging error.
[0042] In combination with the first aspect, in a possible implementation of the first aspect, the plurality of sample digital cultural and educational interaction logs are obtained by downloading initial conversation logs recorded by a conversation interaction server in different text application scenarios;
[0043] Then, the big data push matching network is debugged multiple times in a loop through several sample digital cultural and educational interaction logs, including: multiple debugging example sets constructed based on the several sample digital cultural and educational interaction logs are debugged multiple times in a loop on the big data push matching network, each debugging example set includes two sample digital cultural and educational interaction logs from the same basic digital cultural and educational interaction log but with different textual application scenarios;
[0044] The determining of the network debugging error based on the prior subscription content push tags and the directional subscription content push tags of the plurality of sample digital cultural and educational interaction logs comprises: determining the network debugging error based on the prior subscription content push tags and the directional subscription content push tags of the sample digital cultural and educational interaction logs in each debugging example set.
[0045] In combination with the first aspect, in a possible implementation of the first aspect, determining the network debugging error based on a priori subscription content push tags and directional subscription content push tags of sample digital cultural and educational interaction logs in each debugging example set includes:
[0046] Determine the push decision training error based on the prior subscription content push label and the targeted subscription content push label of each sample digital cultural and educational interaction log;
[0047] Based on the prior subscription content push tags of two sample digital cultural and educational interaction logs in each debugging example set, determine the push sorting prior annotation of the debugging example set, and based on the linkage combination result of the directional subscription content push tags of the two sample digital cultural and educational interaction logs and the push sorting prior annotation, determine the push sorting training error;
[0048] Determine the push induction training error based on the linkage combination results between the prior subscription content push tags and the targeted subscription content push tags of each sample digital cultural and educational interaction log;
[0049] The network debugging error is determined based on the push decision training error, the push collation training error and the push induction training error.
[0050] In combination with the first aspect, in a possible implementation of the first aspect, determining the network debugging error based on a priori subscription content push tags and directional subscription content push tags of sample digital cultural and educational interaction logs in each debugging example set includes:
[0051] Determine a first tag variable of a priori subscription content push tag and a second tag variable of a directional subscription content push tag of the input sample digital cultural and educational interaction log;
[0052] For each sample digital cultural and educational interaction log, determining a first linkage combination result between a priori subscription content push tag and the first tag variable, and a second linkage combination result between a directional subscription content push tag and the second tag variable;
[0053] Based on the obtained first linkage combination results and second linkage combination results, determining the push dynamic update training error of the input sample digital cultural and educational interaction log;
[0054] The network debugging error is determined based on the pushed dynamically updated training error.
[0055] In a second aspect, the present application also provides a big data push matching system, including: a memory for storing program instructions and data; a processor for coupling with the memory and executing instructions in the memory to implement the method as described above.
[0056] In a third aspect, the present application also provides a computer storage medium comprising instructions, which implement the above method when executed on a processor.
[0057] In order to solve the above problems, this application proposes a personalized matching method for digital cultural and educational resources based on a big data push matching system. The method first analyzes the digital cultural and educational interaction log, identifies the conversation state style within each analysis cycle, and then obtains the user's language habit description. This step helps the system to understand the user's personality and interests more deeply, laying the foundation for subsequent accurate push.
[0058] Secondly, this application collects digital cultural and educational texts from digital cultural and educational interaction logs, and performs conversation text semantic mining on these texts to obtain the conversation text embedding semantics corresponding to each text. By capturing the deep semantic connections between texts, the system can more accurately understand the user's true intentions and interests.
[0059] Furthermore, based on the obtained user language habit description and conversation text embedding semantics, this application constructs conversation interest linear knowledge representation and user push preference knowledge representation respectively. These two knowledge representations depict the user's interests and preferences from different dimensions, providing more comprehensive information support for the system's recommendation decisions.
[0060] Finally, this application obtains information recommendation decision-making assistance knowledge by performing knowledge cross-operation on user push preference knowledge representation and conversation interest linear knowledge representation, and determines the directional subscription content push label of the digital cultural and educational interaction log based on this knowledge. This method not only improves the accuracy and personalization of push, but also helps users discover more content resources that meet their interests and needs. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] The accompanying drawings herein are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the description, are used to explain the principles of the present application.
[0062] Figure 1 It is a flowchart of a personalized matching method for digital cultural education resources based on AIGC provided in an embodiment of the present application.
[0063] Figure 2 It is a structural block diagram of a big data push matching system 300 provided in an embodiment of the present application. DETAILED DESCRIPTION
[0064] The technical solution in this application will be described below in conjunction with the accompanying drawings.
[0065] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application.
[0066] It should be noted that the terms "first", "second", etc. in the specification of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.
[0067] The method embodiments provided in the embodiments of the present application can be executed in a big data push matching system, a computer device or a similar computing device. Taking the operation on a big data push matching system as an example, the big data push matching system may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory for storing data. Optionally, the above-mentioned big data push matching system may also include a transmission device for communication functions. It can be understood by those of ordinary skill in the art that the above-mentioned structure is only for illustration and does not limit the structure of the above-mentioned big data push matching system. For example, the big data push matching system may also include more or fewer components than those shown above, or have a configuration different from that shown above.
[0068] The memory can be used to store computer programs, for example, software programs and modules of application software, such as a computer program corresponding to a personalized matching method for digital cultural education resources based on AIGC in an embodiment of the present application. The processor executes various functional applications and data processing by running the computer program stored in the memory, that is, implementing the above method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the big data push matching system via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0069] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the big data push matching system. In one example, the transmission device includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.
[0070] Based on this, see Figure 1 , Figure 1 It is a flow chart of a personalized matching method for digital cultural education resources based on AIGC provided in an embodiment of the present application. The method is applied to a big data push matching system and may further include steps 110 to 140.
[0071] Step 110 , when the big data push matching system parses the digital cultural and educational interaction log, it obtains the user language habit description of the corresponding parsing period according to the conversation state style identified in each parsing period.
[0072] Step 120: The big data push matching system collects digital cultural and educational texts from the digital cultural and educational interaction logs, and performs conversation text semantic mining on each collected digital cultural and educational text to obtain conversation text embedded semantics corresponding to each digital cultural and educational text.
[0073] Step 130: The big data push matching system obtains the corresponding linear knowledge representation of conversation interests based on the obtained language habit descriptions of each user, and obtains the corresponding knowledge representation of user push preferences based on the obtained embedded semantics of each conversation text.
[0074] Step 140: The big data push matching system performs a knowledge cross-operation on the user push preference knowledge representation and the conversation interest linear knowledge representation to obtain corresponding information recommendation auxiliary decision-making knowledge, and determines a targeted subscription content push tag for the digital cultural and educational interaction log based on the information recommendation auxiliary decision-making knowledge.
[0075] The following is a detailed introduction to the technical solution for implementing the big data push matching system, using a specific application scenario example to illustrate how the solution is implemented.
[0076] Application scenario: content push system for online education platform
[0077] Online education platforms generate a large number of interaction logs between users and digital cultural and educational content every day. In order to improve user experience and content targeting, the platform uses a big data push matching system to optimize content push. The big data push matching system first parses these interaction logs. During the parsing process, the system obtains the user language habit description of the corresponding parsing cycle based on the conversation state style identified in each parsing cycle (such as user activity, question frequency, vocabulary used, etc.). For example, if a user asks questions many times and uses professional terms in the parsing cycle, the system can describe his language habits as "professional and inquisitive". Next, the system collects digital cultural and educational texts from the interaction logs, which may include course descriptions, user comments, question answers, etc. The system performs conversation text semantic mining on each collected digital cultural and educational text, and obtains the conversation text embedding semantics corresponding to these texts through natural language processing technology (such as word embedding, deep learning model, etc.). These embedding semantics can capture the deep meaning and contextual relationship of the text. Based on the obtained user language habit description, the system further generates the corresponding conversation interest linear knowledge representation. These representations can quantify the user's interest preferences and conversation style. At the same time, the system will also generate user push preference knowledge representations based on the acquired conversation text embedding semantics. These representations reflect the user's preference for different types of content. Finally, the system will perform knowledge cross-operation on the user push preference knowledge representation and the conversation interest linear knowledge representation. This process aims to integrate two different types of knowledge to obtain a more comprehensive and accurate user portrait. Through this cross-operation, the system is able to generate corresponding information recommendation auxiliary decision-making knowledge, which will be used to guide subsequent targeted content push. Finally, the system will determine the targeted subscription content push tags of the digital cultural and educational interaction log based on this information recommendation auxiliary decision-making knowledge, such as "recommended for professional users interested in AI" or "introductory courses suitable for beginners". These tags will ensure that users receive content recommendations that are highly matched to their interests and needs.
[0078] Furthermore, an example of the application of the big data push matching system in the personalized matching of digital cultural education resources based on AIGC is given.
[0079] In this application scenario, the big data push matching system is the core component for realizing personalized matching of digital cultural and educational resources. It deeply analyzes the interaction logs between users and digital cultural and educational resources, combined with AIGC technology, to provide users with customized learning content and recommendations. The following is the detailed workflow of the system: The system first regularly collects and analyzes digital cultural and educational interaction logs. When parsing these logs, the system pays special attention to the conversation state style within each parsing cycle. These styles may include the user's speaking frequency, word usage habits, questioning patterns, etc. By analyzing these conversation styles, the system can accurately form a description of the user's language habits within the corresponding parsing cycle. These descriptions provide basic data for subsequent personalized matching; after completing the analysis of the user's language habits, the system will further collect digital cultural and educational texts from the interaction logs. These texts may include user comments, feedback, course content, discussion topics, etc. For each collected text, the system will conduct in-depth conversation text semantic mining. This process aims to understand the meaning, emotion, and intention behind the text, and then obtain the conversation text embedding semantics corresponding to each text. These semantic information can more comprehensively reflect the user's needs and preferences for digital cultural education resources; based on the user's language habit description obtained in step 110, the system can further derive the user's conversation interest linear knowledge representation. These representations show the user's interest in different types of topics and content in a quantitative form. At the same time, combined with the conversation text embedded semantics obtained in step 120, the system can also generate user push preference knowledge representation. These representations reflect the user's preferences and needs in the learning process, providing an important basis for subsequent personalized push; after obtaining the user's conversation interest linear knowledge representation and push preference knowledge representation, the system will perform knowledge cross-operation. The purpose of this step is to integrate two different types of knowledge to form a more comprehensive and accurate user portrait. Through this cross-operation, the system can generate information recommendation auxiliary decision-making knowledge, which comprehensively considers factors such as user interests, needs and learning habits. Finally, based on these auxiliary decision-making knowledge, the system will determine a series of targeted subscription content push tags for each user. These tags ensure that users can receive digital cultural education resource recommendations that are highly matched with their personalized needs.
[0080] In general, through this series of technical solutions and processes, the big data push matching system can effectively realize personalized matching of digital cultural education resources based on AIGC. It can not only deeply understand the needs and preferences of users, but also provide users with accurate and personalized learning content recommendations, thereby significantly improving the effect of digital cultural education and user experience.
[0081] Next, the technical terms involved in the above steps 110 to 140 are explained and given examples.
[0082] Digital cultural and educational interactive log:
[0083] Explanation: Digital cultural and educational interaction logs refer to records of interactions between users and systems or content in digital cultural and educational environments. These logs usually contain user behavior data, timestamps, interaction types, content details, and other information, and are used to analyze user learning habits, preferences, and system performance.
[0084] For example: On an online education platform, user Zhang San paused, fast-forwarded, took notes, and performed other operations while watching a history course video. These operations as well as the related time points, content details, etc. will be recorded in the digital cultural and educational interaction log.
[0085] Parsing cycle:
[0086] Explanation: The analysis period refers to a fixed time period set when performing data analysis or processing. During this period, the system will collect and process relevant data to obtain user behavior patterns, trends or other relevant information within a certain period.
[0087] For example: A big data push matching system may set an analysis cycle of 24 hours, that is, the system will collect and analyze all digital cultural and educational interaction logs in the past 24 hours to obtain the user's latest behavior patterns and preferences.
[0088] Session state style:
[0089] Explanation: Conversational state style refers to the unique behavior and language patterns exhibited by users when interacting with the system or content. These patterns can include the user's speaking frequency, word choice, emotional expression, etc., reflecting the user's personal style, emotional state, and interaction habits.
[0090] For example, when using the discussion area of an online education platform, user Li Si often posts long speeches, uses precise and professional words, and demonstrates a serious and rigorous conversational style.
[0091] User language habit description:
[0092] Explanation: User language habit description refers to the summary or representation of a user's language habits obtained by analyzing the user's language usage pattern. These descriptions may include the user's vocabulary selection, grammatical structure, sentence preference, etc., which are used to reveal the user's language usage characteristics and style.
[0093] For example, after analyzing the speech of user Wang Wu on the online education platform, it is found that he often uses simple and clear sentence structures and likes to use vivid metaphors and figurative descriptions. Therefore, Wang Wu's language habits can be described as "concise and vivid". This description helps the system understand the user's needs and preferences more accurately, thereby providing more personalized content recommendations.
[0094] Digital cultural and educational texts:
[0095] Explanation: Digital cultural and educational texts refer to textual content produced in a digital cultural and educational environment. These texts usually contain educational information, course content, user comments, discussion topics, etc. Digital cultural and educational texts are an important part of digital cultural and educational resources, providing users with learning materials and information.
[0096] For example: a course description about literature on an online education platform, users’ evaluations of the course, comments in the course discussion area, etc. can all be considered digital cultural and educational texts.
[0097] Conversational text semantic mining:
[0098] Explanation: Conversation text semantic mining refers to the use of natural language processing technology to conduct in-depth analysis of conversation texts to extract and understand the semantic information in the text. This includes identifying the topic, sentiment, intent, entity, etc. of the text, as well as the association and contextual relationship between texts.
[0099] For example, in the discussion area of an online education platform, conversations between users may include understanding of course content, sharing of learning experiences, answering questions, etc. Through conversation text semantic mining, the system can identify the topics and key information of these conversations, such as learning difficulties or points of interest mentioned by a user.
[0100] Conversation text embedding semantics:
[0101] Explanation: Conversation text embedding semantics refers to the process of converting conversation text into low-dimensional dense vectors that can capture the semantic information of the text. Through embedding techniques (such as word embedding, sentence embedding, etc.), text can be converted into a mathematical representation that computers can understand and operate.
[0102] For example, for user comments on an online education platform, the system can use word embedding technology to convert each word into a vector representation, and further combine these word vectors into an embedded representation of a sentence or text. These embedded semantics can be used to compare the similarity between different texts, perform text classification or clustering, and other tasks.
[0103] Conversational interest linear knowledge representation:
[0104] Explanation: Conversation interest linear knowledge representation refers to a linear model built based on the user's conversation data (such as speech content, frequency, etc.) to represent the user's interest in different topics or content. This representation can quantify the user's interest in a specific field and help the system make personalized recommendations.
[0105] For example, by analyzing the user's learning records and forum posts on the online education platform, the system can find that the user has a higher interest in history and a lower interest in mathematics. These interest levels can be quantified through linear knowledge representation to guide subsequent recommendation algorithms.
[0106] Knowledge representation of user push preferences:
[0107] Explanation: User push preference knowledge representation refers to the knowledge representation built based on the user's historical behavior, preference settings and other information, which is used to describe the user's preferences and needs for push content. This representation can help the system understand the user's expectations more accurately, thereby providing personalized push that meets the user's needs.
[0108] For example, users of online education platforms may set preference tags (such as liking video courses and disliking audio courses), and their learning behaviors also reflect their preferences (such as frequently watching technology video courses). The system can construct a knowledge representation of user push preferences based on this information and behavior to ensure that the content pushed to users meets their preferences and needs.
[0109] Knowledge cross-operation:
[0110] Explanation: Knowledge cross-operation refers to the fusion, cross-operation or combination of knowledge from different sources, types or levels to generate new, richer or more valuable knowledge. In personalized recommendation systems, this usually involves crossing user interests and behavior data with content metadata, external knowledge bases, etc. to discover new associations, patterns or recommendation opportunities.
[0111] For example, in an online education platform, knowledge cross-operation can cross the linear knowledge representation of user conversational interests (reflecting the user's interest in different topics) with the knowledge representation of user push preferences (reflecting the user's preference for push content) to generate a more comprehensive user profile. Such cross-operation can help the system discover new areas or content types that users may be interested in but have not yet come into contact with, and thus provide more accurate personalized recommendations.
[0112] Information recommendation to assist decision-making knowledge:
[0113] Explanation: Information recommendation decision-making knowledge refers to a series of rules, models, algorithms and data sets that are relied upon when making personalized recommendations, which together form the decision-making basis of the recommendation system. This knowledge can help the system understand user needs, preferences and behavior patterns, thereby making smarter recommendation decisions.
[0114] For example, the recommendation system of an online education platform may use the user's historical learning data, metadata of course content, user feedback information, etc. as information recommendation auxiliary decision-making knowledge. By analyzing this data, the system can identify the user's learning style, progress and difficulty, and then recommend courses, learning materials or learning paths suitable for the user. Such recommendation decision-making knowledge helps improve the user's learning efficiency and satisfaction.
[0115] Targeted subscription content push tags:
[0116] Explanation: Targeted subscription content push tags are tags generated based on user interests, preferences, and behavior data, used to identify and classify content that users may be interested in. These tags can help the system more accurately locate user needs, thereby achieving personalized content push.
[0117] For example, users of online education platforms may be interested in learning different types of courses, such as programming, design, language learning, etc. The system can label users with corresponding targeted subscription content push tags based on their learning history and preference settings, such as "Python programming enthusiasts", "design beginners", etc. When new related courses or resources are launched, the system can accurately push content to the corresponding user groups based on these tags.
[0118] The step 140 is further described below.
[0119] From the perspective of personalized matching of digital cultural and educational resources based on AIGC, the core task of the big data push matching system is to use advanced data analysis technology, combined with users' conversational interests and behavioral preferences, to push highly personalized digital cultural and educational resources to users. The following is a detailed introduction on how the system uses the knowledge representation of user push preferences and the linear knowledge representation of conversational interests to perform knowledge cross-operation, thereby obtaining information recommendation auxiliary decision-making knowledge and ultimately determining the targeted subscription content push tags.
[0120] Knowledge cross-operation: The big data push matching system will first construct the user push preference knowledge representation and the conversation interest linear knowledge representation respectively. The user push preference knowledge representation is mainly based on the user's historical behavior data, preference settings, and feedback information, etc., reflecting the user's overall preference for the pushed content. The conversation interest linear knowledge representation quantifies the user's interest in specific topics or content by analyzing the user's conversation data, such as speaking frequency, content topics, etc. Next, the system will perform a knowledge cross-operation to merge the two knowledge representations. The purpose of this step is to discover the potential correlation and complementarity between user interests and push preferences. Through cross-operation, the system can generate a more comprehensive and detailed user portrait, which not only helps to understand the user's current needs, but also predict the user's future interest trends.
[0121] Information recommendation decision-making support knowledge: The result of knowledge cross-operation is a series of information recommendation decision-making support knowledge. This knowledge exists in the form of rules, models or algorithms, which together form the decision-making basis of the recommendation system. For example, the system may find that the user prefers learning resources in the form of videos and has a strong interest in history courses. Based on this decision-making support knowledge, the system can perform content screening and recommendation more targetedly.
[0122] Determine targeted subscription content push tags: Ultimately, the big data push matching system will use information recommendation auxiliary decision-making knowledge to determine the targeted subscription content push tags for digital cultural and educational interaction logs. These tags are generated based on the user's personalized needs and preferences, and are used to identify and classify content that the user may be interested in. For example, the system may label the user as "history lover" or "video learning preference".
[0123] When new digital cultural and educational resources are launched, the system will accurately push the content to the corresponding user groups based on these tags. In this way, users can not only obtain the content they are interested in in a timely manner, but also explore more learning resources in related fields under the guidance of the system, thus achieving a more efficient and personalized learning experience.
[0124] The beneficial effects of this application are mainly reflected in the following aspects:
[0125] Improve the accuracy of personalized push notifications: By analyzing digital cultural and educational interaction logs and obtaining user language habit descriptions based on the session state style identified in each analysis cycle, the push notification system can more deeply understand the user's personalized needs and preferences. This meticulous analysis helps the system generate more accurate user portraits, thereby providing users with more personalized content recommendations.
[0126] Enhanced semantic understanding capability: This application mines the conversation text semantics of digital cultural and educational texts to obtain the conversation text embedding semantics corresponding to each digital cultural and educational text. This semantic level of understanding enables the system to capture the user's true intentions and interests, providing strong support for subsequent precise push notifications.
[0127] Enriching knowledge representation dimensions: Based on the obtained user language habit description and conversation text embedding semantics, this application further generates conversation interest linear knowledge representation and user push preference knowledge representation. These two knowledge representations depict user interests and preferences from different dimensions, providing more comprehensive information for the system's recommendation decisions.
[0128] Optimize the recommendation decision process: By performing knowledge cross-operation on the user push preference knowledge representation and the session interest linear knowledge representation, this application obtains information recommendation decision-making auxiliary knowledge. This cross-fusion method helps to discover the potential connection between user interests and push preferences, thereby optimizing the recommendation decision process and improving the accuracy and satisfaction of recommendations.
[0129] Realize accurate content push: Based on the knowledge of information recommendation and decision-making assistance, this application determines the targeted subscription content push tags for digital cultural and educational interactive logs. These tags can accurately identify the content types that users are interested in, so that the system can push the right content to the right users, realizing the real sense of accurate push.
[0130] To sum up, this application has significantly improved the accuracy and efficiency of personalized matching of digital cultural and educational resources by comprehensively using technical means such as big data analysis and natural language processing, providing users with a better and more personalized learning experience.
[0131] In some possible embodiments, the method further includes: determining whether the number of digital cultural and educational texts collected during the target subscription content push phase reaches a preset number threshold.
[0132] Then, based on the obtained descriptions of the language habits of each user, a corresponding linear knowledge representation of conversation interest is obtained, and based on the obtained embedded semantics of each conversation text, a corresponding knowledge representation of user push preference is obtained, including: when the preset number threshold is reached, based on the language habit descriptions of each user obtained during the target subscription content push stage, a corresponding linear knowledge representation of conversation interest is obtained, and based on the embedded semantics of each conversation text obtained during the target subscription content push stage, a corresponding knowledge representation of user push preference is obtained.
[0133] Based on the information recommendation auxiliary decision-making knowledge, determining the targeted subscription content push tag of the digital cultural and educational interaction log includes: based on the information recommendation auxiliary decision-making knowledge, determining the targeted subscription content push tag of the digital cultural and educational interaction log in the target subscription content push stage.
[0134] In some possible embodiments, the method performed by the big data push matching system further includes a key step: determining whether the number of digital cultural and educational texts collected during the target subscription content push phase reaches a preset number threshold. This step is crucial to ensure the accuracy and personalization of the pushed content.
[0135] Specifically, when the number of digital cultural and educational texts collected by the system during the target subscription content push phase reaches a preset threshold, it will trigger a series of subsequent operations. This preset threshold is set based on system requirements and experience to ensure that there is enough data to generate accurate user portraits and recommendation decisions.
[0136] Once the preset number threshold is reached, the system will generate the corresponding conversation interest linear knowledge representation based on the language habit descriptions of each user obtained during the target subscription content push phase. This process involves in-depth analysis and understanding of user language habits to capture the user's personality and interest characteristics. Through the conversation interest linear knowledge representation, the system can more accurately grasp the user's preference for different types of content.
[0137] At the same time, the system will also generate corresponding user push preference knowledge representation based on the embedded semantics of each conversation text obtained during the target subscription content push phase. Conversation text embedded semantics is the deep meaning of the text extracted through deep learning and natural language processing technology, which can reflect the user's interest and attitude towards specific topics or content. The user push preference knowledge representation is built based on these embedded semantics to describe the user's preferences and needs for push content.
[0138] After obtaining the linear knowledge representation of conversation interests and the knowledge representation of user push preferences, the system will perform knowledge cross-operation to generate information recommendation decision-making support knowledge. This process involves fusing and cross-analyzing the two knowledge representations to discover the potential connections and rules between user interests and push preferences. Information recommendation decision-making support knowledge provides comprehensive and accurate information support for the system's recommendation decisions.
[0139] Finally, based on the generated information recommendation auxiliary decision-making knowledge, the system will determine the targeted subscription content push tags of the digital cultural and educational interaction logs in the target subscription content push stage. These tags are generated based on the user's interests and preferences, and are used to identify and classify the types of content that the user may be interested in. Through targeted subscription content push tags, the system can accurately push appropriate content to the corresponding user groups, realizing personalized content recommendation services.
[0140] In general, this technical solution uses big data analysis, natural language processing and other technical means to achieve in-depth analysis and understanding of digital cultural and educational interaction logs, providing users with more accurate and personalized content recommendation services. At the same time, by setting preset number thresholds and triggering subsequent operations, the accuracy and reliability of pushed content are ensured.
[0141] Under some preferred design ideas, conversation text semantic mining is performed on each collected digital cultural text to obtain the conversation text embedding semantics corresponding to each digital cultural text, including: for each digital cultural text, the following processing is implemented: for a digital cultural text, text word vector mining is performed on the digital cultural text through a deep residual model node to obtain the original text word vector relationship network of the digital cultural text; the original text word vector relationship network is input into several cascaded vector relationship network processing nodes to obtain the final text word vector relationship network of the last vector relationship network processing node, and based on the final word vector relationship network, the final word vector relationship network of the last vector relationship network processing node is obtained. The final text word vector relationship network obtains the embedded semantics of the conversation text; wherein, the processing process of each vector relationship network processing node includes upsampling the input text word vector relationship network to obtain a first transition text word vector relationship network, and performing an expansion sliding filtering operation on the first transition text word vector relationship network to obtain a second transition text word vector relationship network, and performing feature focusing-based processing on the second transition text word vector relationship network to obtain a third transition text word vector relationship network, and performing feature optimization on the third transition text word vector relationship network and the input text word vector relationship network to obtain the final text word vector relationship network.
[0142] Under some optimal design ideas, the big data push matching system deeply processes the digital cultural and educational texts in the digital cultural and educational interaction logs to obtain the embedded semantics of its conversational texts. This process involves multiple refined steps to ensure the accuracy and depth of semantic mining.
[0143] First, for each digital cultural and educational text collected, the system will mine the text word vector through the deep residual model node. The deep residual model is a commonly used model in deep learning, which can effectively solve the problems of gradient disappearance and model degradation in the training process of deep neural networks. Through this model, the system can extract the key information in the text and convert it into the original text word vector relationship network. This word vector relationship network captures the association and semantic information between words in the text.
[0144] Next, this original text word vector relationship network will be entered into several cascaded vector relationship network processing nodes. Each processing node will perform a series of processing operations on the input text word vector relationship network, including upsampling, expansion sliding filtering, and feature-focused processing. Upsampling processing can increase the dimension of the data, allowing the model to learn more detailed information; expansion sliding filtering is a special convolution operation that can expand the receptive field and capture more contextual information without increasing the complexity of the model; and feature-focused processing enhances important features, allowing the model to pay more attention to features that affect the results.
[0145] After processing at each processing node, the system will obtain a final text word vector relationship network. This relationship network not only retains the key information in the original text, but also enhances the expressiveness of features through a series of optimization operations. Based on this final text word vector relationship network, the system can further extract the embedded semantics of the conversation text, that is, the representation of the text in the semantic space.
[0146] It is worth noting that the design of the cascaded vector relationship network processing nodes enables the entire processing process to abstract and refine text information layer by layer, ensuring the depth and accuracy of semantic mining. At the same time, the application of technologies such as deep residual model and expansion sliding filter also ensures the stability and robustness of the model when processing complex and diverse digital cultural and educational texts.
[0147] In summary, through the application of this optimal design idea and technical means, the big data push matching system can accurately and deeply mine the conversational text embedding semantics of digital cultural and educational texts, providing a solid data foundation for subsequent personalized recommendations.
[0148] In some other design ideas, for a digital cultural and educational text, text word vector mining is performed on the digital cultural and educational text through a deep residual model node to obtain an original text word vector relationship network of the digital cultural and educational text, including: obtaining dialogue information corresponding to a user emotion description dimension from dialogue information of several description dimensions corresponding to the digital cultural and educational text, the several description dimensions including a user emotion description dimension and a sentence logic description dimension; through the deep residual model node, text word vector mining is performed on the dialogue information corresponding to the user emotion description dimension to obtain the original text word vector relationship network.
[0149] In other design ideas, the big data push matching system uses a more refined and targeted method when processing digital cultural and educational texts to obtain their original text word vector relationship network. This method pays special attention to the user emotion description dimension in the text to more accurately capture the user's emotional tendencies and interests.
[0150] First, the system will specifically extract the dialogue information corresponding to the user emotion description dimension from several description dimensions corresponding to the digital cultural and educational text. These description dimensions are pre-defined by the system and are used to characterize the characteristics of the text from different angles. Among them, the user emotion description dimension mainly focuses on the user emotions expressed in the text, such as joy, sadness, anger, etc.; while the word and sentence logic description dimension focuses more on the word and sentence structure and logical relationship of the text.
[0151] After obtaining the conversation information corresponding to the user's emotion description dimension, the system will use the deep residual model node to mine text word vectors for this information. The deep residual model is a powerful deep learning model that can effectively solve the gradient vanishing and representation bottleneck problems in deep neural networks by learning the residuals in the data (that is, the difference between the actual value and the predicted value). Through this model, the system can deeply mine the key information and semantic relationships in the user's emotion description and convert them into word vectors.
[0152] These word vectors not only capture the key words and phrases in the user's emotional description, but also reflect the semantic connection and contextual information between words through the spatial distribution and mutual relationship of the vectors. In this way, the original text word vector relationship network forms a rich and accurate user emotion expression space, providing strong support for subsequent semantic mining and personalized recommendations.
[0153] It is worth noting that this processing method that focuses on the dimension of user emotion description enables the system to more accurately grasp the user's emotional tendencies and interests. In the field of digital culture and education, the user's emotional state is often closely related to his or her acceptance of content and interest preferences. Therefore, by deeply mining and analyzing user emotion description information, the system can more accurately understand the user's needs and preferences, thereby providing more personalized and accurate content recommendation services.
[0154] Furthermore, based on the obtained embedded semantics of each conversation text, the corresponding user push preference knowledge representation is obtained by any one of the following methods: integrating the embedded semantics of each conversation text to obtain the user push preference knowledge representation; averaging the embedded semantics of each conversation text to obtain the user push preference knowledge representation; down-sampling the embedded semantics of each conversation text to obtain the user push preference knowledge representation.
[0155] Furthermore, in the big data push matching system, based on the obtained embedded semantics of each conversation text, the system can adopt a variety of methods to obtain the corresponding user push preference knowledge representation, including but not limited to integration processing, average processing and downsampling processing.
[0156] First, integration processing is a method of fusing the semantic embeddings of multiple conversation texts into a unified representation. Specifically, the system can perform weighted averaging, maximum pooling, or concatenation operations on the semantic embeddings of each conversation text to capture the common interests and preferences of users in multiple conversations. Through integration processing, the system can integrate the information of multiple conversations to form a comprehensive and accurate knowledge representation of user push preferences.
[0157] Secondly, averaging is a simple and effective way to obtain knowledge representation of user push preferences. The system can directly average the semantic embeddings of each session text to obtain a comprehensive representation. This comprehensive representation reflects the average interests and preferences of users in multiple sessions. Although averaging may ignore special information in some sessions, it can provide a relatively stable knowledge representation of user push preferences.
[0158] Finally, downsampling is a method of selecting representative samples from multiple conversation text embeddings. The system can use random downsampling, max pooling downsampling, or attention-based downsampling to select the most representative samples from each conversation text embedding. These representative samples can reflect the user's main interests and preferences in different conversations. Through downsampling, the system can reduce computational complexity while retaining key information about the user's push preferences.
[0159] It should be noted that the above three methods are just a few common methods for obtaining user push preference knowledge representation, and they are not the only options. In practical applications, the system can choose the most suitable method to obtain user push preference knowledge representation according to specific needs and scenarios. In addition, in order to further improve the accuracy and reliability of the representation, the system can also combine multiple methods for comprehensive processing to obtain a more comprehensive and detailed user push preference knowledge representation.
[0160] In summary, the big data push matching system can obtain the corresponding user push preference knowledge representation based on the embedded semantics of each conversation text through integration processing, averaging processing or downsampling processing. These representations can accurately reflect the user's interests and preferences, and provide strong support for subsequent personalized recommendations.
[0161] In other possible embodiments, when parsing the digital cultural and educational interaction log, a user language habit description of the corresponding parsing cycle is obtained according to the session state style identified in each parsing cycle, including: for each parsing cycle, the following operations are performed: based on the difference between the digital cultural and educational text parsed in one parsing cycle and the digital cultural and educational text parsed in at least one past parsing cycle closest to the one parsing cycle, whether a semantic fault occurs in the one parsing cycle is determined; based on the judgment point of view, the one parsing cycle is marked to obtain the corresponding user language habit description; wherein, when the judgment point of view is continuous, the first language habit variable is marked, and when the judgment point of view is semantic fault, the second language habit variable is marked.
[0162] In other possible embodiments, when processing digital cultural and educational interaction logs, the big data push matching system will obtain the language habit description of the corresponding user based on the conversation state style identified in each analysis cycle. This process aims to more deeply understand the user's communication style and expression habits, so as to provide personalized content that better meets their needs.
[0163] Specifically, for each parsing cycle, the system performs a series of operations to analyze the user's language habits. First, the system determines whether a semantic gap has occurred in the parsing cycle based on the difference between the digital cultural and educational text parsed in the current parsing cycle and the digital cultural and educational text parsed in the most recent past parsing cycle. A semantic gap refers to a significant change or incoherence in the semantics or theme of a text during a user's continuous communication process.
[0164] In order to accurately determine the occurrence of semantic faults, the system can use a variety of natural language processing technologies, such as text similarity calculation, topic model analysis, keyword extraction, etc. Through these technologies, the system can quantitatively analyze the differences between texts to determine whether a semantic fault has occurred.
[0165] Next, the system will mark each parsing cycle based on the judgment results to obtain the corresponding description of the user's language habits. When the judgment point is "continuous", that is, the user maintains relatively consistent semantics and topics in the continuous communication process, the system will mark the first language habit variable. This means that the user has stable and coherent language habits in communication.
[0166] On the contrary, when the judgment point of view is "semantic fault", that is, the user has a large semantic or topic change in the continuous communication process, the system will mark the second language habit variable. This means that the user has diverse and changing language habits in communication.
[0167] In some preferred embodiments, based on the obtained language habit descriptions of each user, a corresponding linear knowledge representation of conversation interest is obtained, including: sorting the obtained language habit descriptions of each user according to the priority between the corresponding parsing cycles to obtain a user language habit description sequence; based on preset description variable mapping features, determining the keyword mapping vectors corresponding to each of the multiple interest keywords from the user language habit description sequence; based on the keyword mapping vectors of each of the multiple interest keywords, obtaining the linear knowledge representation of conversation interest.
[0168] In some preferred embodiments, the big data push matching system will use the obtained descriptions of each user's language habits to further generate corresponding linear knowledge representations of conversation interests. This process involves multiple steps, aiming to extract the user's core interests from their language habits and present them in the form of linear knowledge representations to facilitate subsequent recommendation processing.
[0169] First, the system will sort out the obtained user language habit descriptions according to the priority between the corresponding analysis cycles to form an orderly sequence of user language habit descriptions. This sorting process ensures the coherence and timing of the habit descriptions, so that subsequent analysis can be carried out according to the user's communication timeline.
[0170] Next, the system will identify multiple interest keywords from the user's language habit description sequence based on the preset description variable mapping features. These keywords are core words that reflect the user's interests and preferences, and are crucial for understanding the user's conversational intent and content. The description variable mapping feature is a set of preset rules or models that are used to convert the user's language habit description into a quantifiable feature vector.
[0171] After determining the keywords of interest, the system will generate a corresponding keyword mapping vector for each keyword. This process usually involves word vector technology, such as Word2Vec or GloVe, which can convert words into vector representations in high-dimensional space and capture the semantic relationship and similarity between words. In this way, each keyword of interest is mapped to a specific vector space to form a keyword mapping vector.
[0172] Finally, based on these keyword mapping vectors, the system generates a linear knowledge representation of the conversational interest. This representation is a comprehensive vector that integrates the information of all the interest keywords expressed by the user in the conversation and presents it in a linear manner. This representation not only retains the semantic information of the original keywords, but also facilitates subsequent calculation and processing. In this way, the system can accurately capture the user's interests and preferences, providing strong support for subsequent content recommendations.
[0173] In some possible technical solutions, a knowledge cross-operation is performed on the user push preference knowledge representation and the conversation interest linear knowledge representation to obtain corresponding information recommendation auxiliary decision-making knowledge, and based on the information recommendation auxiliary decision-making knowledge, a targeted subscription content push label of the digital cultural and educational interaction log is determined, including: performing a knowledge dimension adjustment operation on the conversation interest linear knowledge representation so that the knowledge dimension of the conversation interest linear knowledge representation is the same as that of the user push preference knowledge representation; obtaining the information recommendation auxiliary decision-making knowledge based on the linkage combination result between the user push preference knowledge representation and the conversation interest linear knowledge representation; entering the information recommendation auxiliary decision-making knowledge into a push decision tree network to obtain a targeted subscription content push label output by the push decision tree network.
[0174] In some possible technical solutions, the big data push matching system will further process the user push preference knowledge representation and the conversation interest linear knowledge representation to generate information recommendation auxiliary decision-making knowledge, and determine the targeted subscription content push tags of the digital cultural and educational interaction log based on this. The following are the detailed implementation steps.
[0175] First, the system will adjust the knowledge dimension of the linear knowledge representation of conversational interests. This is because the user push preference knowledge representation and the conversational interest linear knowledge representation may be represented on different knowledge dimensions, and direct comparison or combination may lead to information loss or misunderstanding. Therefore, the system needs to adjust the knowledge dimension of the conversational interest linear knowledge representation to be the same as the user push preference knowledge representation to ensure the consistency of the two in subsequent processing.
[0176] The knowledge dimension adjustment operation may include linear transformation, dimensionality reduction or dimensionality increase, depending on the dimensionality of the original knowledge representation and the target dimensionality. For example, if the dimension of the linear knowledge representation of session interests is higher than the knowledge representation of user push preferences, the system may use methods such as principal component analysis (PCA) to reduce the dimensionality; conversely, if the dimensionality is lower, the system may use methods such as zero padding or interpolation to increase the dimensionality.
[0177] Next, the system will obtain information recommendation decision-making assistance knowledge based on the linkage combination results between the adjusted user push preference knowledge representation and the conversation interest linear knowledge representation. The linkage combination here may include weighted average, splicing, dot product or other complex combination methods, aiming to capture the complementarity and correlation between the two knowledge representations. In this way, the system can comprehensively consider the user's long-term preferences and short-term interests to generate a more comprehensive and accurate recommendation decision basis.
[0178] Finally, the system enters the generated information recommendation auxiliary decision knowledge into the push decision tree network. The push decision tree network is a pre-trained machine learning model that can predict the targeted subscription content push tags that best meet user needs based on the input knowledge representation. This model may be built based on algorithms such as decision trees, random forests, or gradient boosting trees, and has powerful classification and regression capabilities. By inputting information recommendation auxiliary decision knowledge, the push decision tree network can output one or more targeted subscription content push tags, which will be used to guide subsequent personalized content push.
[0179] In some optional embodiments, the method is implemented by a big data push matching network, and the debugging method of the big data push matching network includes: performing multiple loop debugging on the big data push matching network through a number of sample digital cultural and educational interaction logs; wherein, in each loop debugging, the following processing is implemented: collecting digital cultural and educational texts from the entered sample digital cultural and educational interaction logs, and performing conversation text semantic mining on each collected digital cultural and educational text to obtain the corresponding conversation text embedding semantics, and, according to the semantic fault state of each parsing cycle in the sample digital cultural and educational interaction log, obtaining the user language habit description of the corresponding parsing cycle; based on the obtained conversation text embedding semantics, obtaining the corresponding user push preference knowledge representation, and based on the obtained user language habit description, obtaining the corresponding conversation interest linear knowledge representation; performing knowledge cross-operation on the user push preference knowledge representation and the conversation interest linear knowledge representation to obtain the corresponding information recommendation decision-making assistance knowledge, and determining the directional subscription content push label based on the information recommendation decision-making assistance knowledge; determining the network debugging error based on the prior subscription content push labels and the directional subscription content push labels of the several sample digital cultural and educational interaction logs, and improving the neural network weight based on the network debugging error.
[0180] In some optional embodiments, the debugging method of the big data push matching network is a complex and delicate process, which aims to ensure that the network can accurately push relevant content based on the user's digital cultural and educational interaction log. The following are the detailed steps of the debugging method:
[0181] First, the system will use several sample digital cultural and educational interaction logs to perform multiple loop debugging on the big data push matching network. These sample logs are collected and labeled in advance, and contain user interaction behaviors and various characteristics in the field of digital culture and education. Through loop debugging, the system can gradually optimize network parameters and improve the accuracy of push.
[0182] In each loop debugging, the system will perform the following processing steps: Collect digital cultural and educational texts from the sample digital cultural and educational interaction logs that have been entered. These texts are generated when users interact with the network and contain information such as user questions, needs, and feedback. Perform conversation text semantic mining on the collected digital cultural and educational texts to obtain the corresponding conversation text embedding semantics. This step aims to extract deep semantic information from the text in order to more accurately understand the user's intentions and needs. According to the semantic fault state of each parsing cycle in the sample digital cultural and educational interaction log, obtain the user language habit description of the corresponding parsing cycle. The semantic fault state reflects the changes in the semantics or themes of users in the continuous communication process, and the user language habit description is a summary and refinement of this change rule. Based on the obtained conversation text embedding semantics, the system further generates the corresponding user push preference knowledge representation. This representation integrates the semantic information of users in multiple conversations and reflects the user's long-term interests and preferences. At the same time, based on the obtained description of each user's language habit, the system will also generate the corresponding conversation interest linear knowledge representation. This representation mainly focuses on the changes in user interests in a single conversation, and captures and describes the user's short-term interests. Next, the system will perform knowledge cross-operation on the user push preference knowledge representation and the conversation interest linear knowledge representation to obtain the corresponding information recommendation decision-making assistance knowledge. This step aims to combine the user's long-term interests and short-term interests to generate a more comprehensive and accurate recommendation basis. Based on the generated information recommendation decision-making assistance knowledge, the system will determine the targeted subscription content push tags. These tags are further refinements and classifications of user needs and interests, and will be used to guide subsequent personalized content push. Finally, the system will determine the network debugging error based on the prior subscription content push tags and targeted subscription content push tags of several sample digital cultural and educational interaction logs. The priori label is the correct result that has been marked in advance, while the targeted label is the predicted result generated by the system based on the current network parameters. By comparing the difference between the two, the system can calculate the network debugging error and improve the neural network weight accordingly. The improved network will capture the user's interests and needs more accurately and improve the accuracy and satisfaction of subsequent push.
[0183] In other optional embodiments, the several sample digital cultural and educational interaction logs are obtained by downloading the initial session logs recorded by the session interaction server in different textual application scenarios; then, the big data push matching network is debugged multiple times in a loop through the several sample digital cultural and educational interaction logs, including: multiple loop debugging of the big data push matching network based on several debugging example sets constructed based on the several sample digital cultural and educational interaction logs, each debugging example set including two sample digital cultural and educational interaction logs from the same basic digital cultural and educational interaction logs but with different textual application scenarios; the network debugging error is determined based on the prior subscription content push tags and the targeted subscription content push tags of the several sample digital cultural and educational interaction logs, including: the network debugging error is determined based on the prior subscription content push tags and the targeted subscription content push tags of the sample digital cultural and educational interaction logs in each debugging example set.
[0184] In other optional embodiments, in order to enhance the generalization ability and adaptability of the big data push matching network, the system uses sample digital cultural and educational interaction logs from different textual application scenarios for debugging. These sample logs are obtained by downloading the initial session logs recorded by the session interaction server, covering a variety of different textual application scenarios, such as online education, digital libraries, cultural exchange platforms, etc.
[0185] When performing multiple cycles of debugging, the system first builds several debugging example sets based on these sample digital cultural and educational interaction logs. Each debugging example set includes two sample digital cultural and educational interaction logs from the same basic digital cultural and educational interaction logs but with different textual application scenarios. This construction method enables the system to fully consider the differences in user behavior and changes in language habits in different application scenarios during the debugging process, thereby improving the accuracy and adaptability of push matching.
[0186] Specifically, the system selects a pair of sample digital cultural and educational interaction logs from each debugging example set as input and reference standards. Then, the system uses the big data push matching network to process the input logs and generate targeted subscription content push tags. Then, the system compares the generated targeted tags with the prior subscription content push tags in the reference standard to calculate the network debugging error.
[0187] In this process, the prior subscription content push label is the correct result that has been marked in advance, representing the actual subscription needs of users in different textual application scenarios. The targeted subscription content push label is the prediction result generated by the system based on the current network parameters and input logs. By comparing the differences between the two, the system can evaluate the performance of the network in different application scenarios and make targeted optimizations and improvements accordingly.
[0188] It should be noted that since user behaviors and language habits may vary greatly in different text-based application scenarios, the impact of these differences on push matching results needs to be fully considered when debugging the network. To this end, the system can use some advanced machine learning technologies, such as transfer learning and domain adaptation, to enhance the network's cross-scenario adaptability.
[0189] In short, through multiple cycles of debugging using these sample digital cultural and educational interaction logs from different textual application scenarios, the system can gradually optimize the parameters and performance of the big data push matching network, and improve its push matching accuracy and adaptability in various application scenarios.
[0190] In some preferred embodiments, the network debugging error is determined based on the prior subscription content push tags and the directional subscription content push tags of the sample digital cultural and educational interaction logs in each debugging example set, including: determining the push decision training error based on the prior subscription content push tags and the directional subscription content push tags of each sample digital cultural and educational interaction log; determining the push collation prior annotation of the debugging example set based on the prior subscription content push tags of two sample digital cultural and educational interaction logs in each debugging example set, and determining the push collation training error based on the linkage combination result of the directional subscription content push tags of the two sample digital cultural and educational interaction logs and the push collation prior annotation; determining the push summary training error based on the linkage combination result between the prior subscription content push tags and the directional subscription content push tags of each sample digital cultural and educational interaction log; and determining the network debugging error based on the push decision training error, the push collation training error and the push summary training error.
[0191] In some preferred embodiments, the system comprehensively evaluates the debugging effect of the big data push matching network through multi-level error calculations. These levels include push decision training error, push collation training error, and push induction training error, which measure the performance of the network from different perspectives.
[0192] First, the system calculates the push decision training error for each sample digital cultural and educational interaction log. This error reflects the accuracy of the network in generating targeted subscription content push labels based on user behavior and language habits. Specifically, the system compares the prior subscription content push label (the correct result marked in advance) and the targeted subscription content push label (the predicted result generated by the network) of each sample log, and calculates the difference between the two as the push decision training error. If the difference is large, it means that the network has deficiencies in push decision-making and needs to be optimized and improved accordingly.
[0193] Secondly, the system determines the push collation training error for each debugging example set. This error takes into account the impact of differences in user behavior and language habits in different textual application scenarios on the push results. Specifically, the system determines the push collation prior annotations for the debugging example set based on the prior subscription content push labels of the two sample digital cultural and educational interaction logs in each debugging example set. This annotation represents the actual subscription needs of users in different application scenarios. Then, the system calculates the push collation training error based on the linkage combination results of the targeted subscription content push labels of the two sample logs (taking into account the prediction results of the two logs) and the push collation prior annotations. If the error is large, it means that the network's adaptability in different application scenarios needs to be improved.
[0194] Finally, the system determines the push induction training error. This error measures the network's ability in inductive learning, that is, whether it can extract commonalities and patterns from multiple sample logs to improve the ability to predict future data. Specifically, the system calculates the push induction training error based on the linkage combination results between the prior subscription content push tags and the targeted subscription content push tags of each sample digital cultural and educational interaction log (taking into account the annotation results and prediction results of multiple logs). If the error is large, it means that the network still has room for improvement in inductive learning.
[0195] Combining the training errors at the above three levels, the system can determine the final network debugging error. This error comprehensively reflects the performance of the big data push matching network in all aspects, and provides a strong basis for subsequent optimization and improvement. Through continuous iteration and optimization, the system can gradually reduce the network debugging error and improve the accuracy and adaptability of the big data push matching network.
[0196] Push decision training error, push collation training error and push induction training error are three important concepts in the debugging process of big data push matching network. They measure the performance of the network from different angles. The following will introduce the meaning and calculation method of these three training errors in detail.
[0197] Push decision training error
[0198] The push decision training error focuses on the accuracy of the network in generating targeted subscription content push labels based on user behavior and language habits. During the training process, the system generates a targeted subscription content push label for each sample digital cultural and educational interaction log. This label is predicted based on the current network parameters and user behavior data. At the same time, each sample log has a prior subscription content push label, which is a pre-annotated correct result that represents the user's actual subscription needs.
[0199] The push decision training error is calculated by comparing the difference between the targeted subscription content push label and the prior subscription content push label. The larger the difference, the lower the accuracy of the network in push decision, and the larger the training error. In order to reduce the push decision training error, the system needs to continuously optimize the network parameters to improve the accuracy of the prediction.
[0200] Push and organize training errors
[0201] The push sorting training error takes into account the impact of differences in user behavior and language habits in different textual application scenarios on push results. During the debugging process, the system will build several debugging example sets, each of which includes two sample digital cultural and educational interaction logs from the same basic digital cultural and educational interaction logs but with different textual application scenarios. The purpose of this design is to enable the network to learn differences in user behavior and changes in language habits in different application scenarios.
[0202] For each debugging example set, the system determines a push collation prior annotation based on the prior subscription content push tags of the two sample logs. This annotation represents the actual subscription needs of users in different application scenarios. Then, the system calculates the push collation training error based on the linkage combination results of the targeted subscription content push tags of the two sample logs (that is, the prediction results of the two logs are comprehensively considered) and the push collation prior annotations. If the error is large, it means that the adaptability of the network in different application scenarios needs to be improved. In order to reduce the push collation training error, the system needs to enhance the cross-scenario adaptability of the network so that it can better adapt to the differences in user behavior and changes in language habits in different textual application scenarios.
[0203] Pushing the inductive training error
[0204] The push inductive training error measures the network's ability to learn inductively. Inductive learning refers to the process of extracting commonalities and regularities from multiple specific examples, which can be used to guide the prediction and decision-making of future data. During the debugging process of the big data push matching network, the system needs to extract commonalities and regularities from multiple sample digital cultural and educational interaction logs to improve the prediction ability of future data.
[0205] The push induction training error is calculated by comparing the linkage combination results between the prior subscription content push tags and the targeted subscription content push tags of each sample digital cultural and educational interaction log. This linkage combination result is obtained by comprehensively considering the annotation results and prediction results of multiple logs, which reflects the network's ability in inductive learning. If the error is large, it means that the network still has room for improvement in extracting commonalities and regularities. In order to reduce the push induction training error, the system needs to improve the network's inductive learning ability so that it can extract more accurate and useful commonalities and regularities from multiple sample logs.
[0206] In other preferred embodiments, the network debugging error is determined based on the prior subscription content push tags and the targeted subscription content push tags of the sample digital cultural and educational interaction logs in each debugging example set, including: determining the first tag variable of the prior subscription content push tags of the entered sample digital cultural and educational interaction logs, and the second tag variable of the targeted subscription content push tags; for each sample digital cultural and educational interaction log, determining the first linkage combination result between the prior subscription content push tags and the first tag variable, and the second linkage combination result between the targeted subscription content push tags and the second tag variable; based on the obtained first linkage combination results and the second linkage combination results, determining the push dynamic update training error of the entered sample digital cultural and educational interaction logs; based on the push dynamic update training error, determining the network debugging error.
[0207] In other preferred embodiments, the system adopts a method based on label variables to determine the network debugging error. This method introduces a first label variable and a second label variable, which respectively represent the variable forms of the prior subscription content push label and the directional subscription content push label, so as to more flexibly handle the differences between different sample digital cultural and educational interaction logs.
[0208] First, the system determines the first label variable of the prior subscription content push label of the sample digital cultural and educational interaction log, and the second label variable of the directional subscription content push label. These label variables can be numerical values, categories, or other forms of identifiers, which are used to represent the corresponding label information in subsequent calculations.
[0209] Next, for each sample digital cultural and educational interaction log, the system determines the first linkage combination result between the prior subscription content push tag and the first tag variable. This linkage combination result reflects the degree of association or difference between the prior subscription content push tag and the first tag variable. Similarly, the system also determines the second linkage combination result between the targeted subscription content push tag and the second tag variable to reflect the association or difference between them.
[0210] After determining the results of each first linkage combination and each second linkage combination, the system will determine the push dynamic update training error of the entered sample digital cultural and educational interaction log based on these results. This error reflects the accuracy of the network when dynamically updating the push content based on user behavior and language habits. Specifically, the system will compare the difference between the first linkage combination results and the second linkage combination results, and calculate the push dynamic update training error. If the difference is large, it means that the network is insufficient in dynamically updating the push content, and corresponding optimization and improvement are needed.
[0211] Finally, the system will determine the network debugging error based on the push dynamic update training error. The network debugging error is an important indicator for comprehensively evaluating the performance of the big data push matching network, which reflects the performance of the network in all aspects. Through continuous iteration and optimization, the system can gradually reduce the network debugging error and improve the accuracy and adaptability of the big data push matching network.
[0212] It should be noted that in actual applications, the specific forms of the first label variable and the second label variable and the method of calculating and pushing the dynamic update training error may vary according to specific needs and scenarios. Therefore, when implementing the technical solution, it is necessary to flexibly adjust and optimize it according to the actual situation.
[0213] Figure 2 The structural block diagram of the big data push matching system 300 is shown, including: a memory 310 for storing program instructions and data; a processor 320 for coupling with the memory 310 and executing the instructions in the memory 310 to implement the above method.
[0214] Furthermore, a computer storage medium is provided, comprising instructions, which implement the above method when executed on a processor.
[0215] In several embodiments provided in the embodiments of the present application, it should be understood that the disclosed device and method can also be implemented in other ways. The device and method embodiments described above are merely schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the device, method and computer program product according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.
[0216] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
[0217] If the function is implemented in the form of a software function module 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 is essentially 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, including several instructions to enable a computer device (which can be a personal computer, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment 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 and other media that can store program code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.
[0218] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A personalized matching method for digital cultural education resources based on AIGC, characterized in that: Applied to a big data push matching system, the method includes: When parsing the digital cultural and educational interaction log, the user language habit description of the corresponding parsing period is obtained according to the conversation state style identified in each parsing period; Collecting digital cultural and educational texts from the digital cultural and educational interaction logs, and performing conversation text semantic mining on each collected digital cultural and educational text to obtain conversation text embedded semantics corresponding to each digital cultural and educational text; Based on the obtained description of each user's language habits, a corresponding linear knowledge representation of conversation interests is obtained, and based on the obtained embedded semantics of each conversation text, a corresponding knowledge representation of user push preferences is obtained. The linear knowledge representation of conversation interests refers to a linear model built based on the user's conversation data, which is used to represent the user's interest in different topics or content. The user push preference knowledge representation is used to describe the user's preference and demand for push content; Performing a knowledge cross-operation on the user push preference knowledge representation and the conversation interest linear knowledge representation to obtain corresponding information recommendation auxiliary decision-making knowledge, and determining a directional subscription content push tag for the digital cultural and educational interaction log based on the information recommendation auxiliary decision-making knowledge; Performing a knowledge cross-operation on the user push preference knowledge representation and the conversation interest linear knowledge representation to obtain corresponding information recommendation decision-making assistance knowledge, and determining a directional subscription content push tag for the digital cultural and educational interaction log based on the information recommendation decision-making assistance knowledge, including: Performing a knowledge dimension adjustment operation on the linear knowledge representation of the conversation interest so that the knowledge dimension of the linear knowledge representation of the conversation interest is the same as the knowledge dimension of the user push preference knowledge representation; Based on the linkage combination result between the user push preference knowledge representation and the session interest linear knowledge representation, the information recommendation auxiliary decision-making knowledge is obtained; Entering the information recommendation auxiliary decision knowledge into a push decision tree network, and obtaining a directional subscription content push tag output by the push decision tree network; The method is implemented by a big data push matching network, and the debugging method of the big data push matching network includes: Through several sample digital cultural and educational interaction logs, the big data push matching network is debugged multiple times in a loop; wherein, in each loop debugging, the following processing is implemented: Collect digital cultural and educational texts from the input sample digital cultural and educational interaction logs, and perform conversation text semantic mining on each collected digital cultural and educational text to obtain the corresponding conversation text embedding semantics, and, according to the semantic fault state of each parsing cycle in the sample digital cultural and educational interaction logs, obtain the user language habit description of the corresponding parsing cycle; Based on the obtained semantic embedding of each conversation text, the corresponding user push preference knowledge representation is obtained, and based on the obtained description of each user's language habits, the corresponding conversation interest linear knowledge representation is obtained; Performing a knowledge cross-operation on the user push preference knowledge representation and the session interest linear knowledge representation to obtain corresponding information recommendation decision-making assistance knowledge, and determining a targeted subscription content push tag based on the information recommendation decision-making assistance knowledge; Based on the prior subscription content push tags and the directional subscription content push tags of the plurality of sample digital cultural and educational interaction logs, a network debugging error is determined, and a neural network weight is improved based on the network debugging error; The sample digital cultural and educational interaction logs are obtained by downloading the initial conversation logs recorded by the conversation interaction server in different text application scenarios; Then, the big data push matching network is debugged multiple times in a loop through several sample digital cultural and educational interaction logs, including: multiple debugging example sets constructed based on the several sample digital cultural and educational interaction logs are debugged multiple times in a loop on the big data push matching network, each debugging example set includes two sample digital cultural and educational interaction logs from the same basic digital cultural and educational interaction log but with different textual application scenarios; The determining of the network debugging error based on the prior subscription content push tags and the directional subscription content push tags of the plurality of sample digital cultural and educational interaction logs comprises: determining the network debugging error based on the prior subscription content push tags and the directional subscription content push tags of the sample digital cultural and educational interaction logs in each debugging example set; Wherein, determining the network debugging error based on the prior subscription content push tags and the directional subscription content push tags of the sample digital cultural and educational interaction logs in each debugging example set includes: Determine the push decision training error based on the prior subscription content push label and the targeted subscription content push label of each sample digital cultural and educational interaction log; Based on the prior subscription content push tags of two sample digital cultural and educational interaction logs in each debugging example set, determine the push sorting prior annotation of the debugging example set, and based on the linkage combination result of the directional subscription content push tags of the two sample digital cultural and educational interaction logs and the push sorting prior annotation, determine the push sorting training error; Determine the push induction training error based on the linkage combination results between the prior subscription content push tags and the targeted subscription content push tags of each sample digital cultural and educational interaction log; Determining the network debugging error based on the push decision training error, the push collation training error and the push induction training error; Wherein, determining the network debugging error based on the prior subscription content push tags and the directional subscription content push tags of the sample digital cultural and educational interaction logs in each debugging example set includes: Determine a first tag variable of a priori subscription content push tag and a second tag variable of a directional subscription content push tag of the input sample digital cultural and educational interaction log; For each sample digital cultural and educational interaction log, determining a first linkage combination result between a priori subscription content push tag and the first tag variable, and a second linkage combination result between a directional subscription content push tag and the second tag variable; Based on the obtained first linkage combination results and second linkage combination results, determining the push dynamic update training error of the input sample digital cultural and educational interaction log; The network debugging error is determined based on the pushed dynamically updated training error.
2. The method according to claim 1, characterized in that The method further includes: determining whether the number of digital cultural and educational texts collected during the target subscription content push phase reaches a preset number threshold; Then, the step of obtaining the corresponding linear knowledge representation of conversation interest based on the obtained language habit descriptions of each user, and obtaining the corresponding knowledge representation of user push preference based on the obtained embedded semantics of each conversation text, includes: when the preset number threshold is reached, obtaining the corresponding linear knowledge representation of conversation interest based on the language habit descriptions of each user obtained in the target subscription content push stage, and obtaining the corresponding knowledge representation of user push preference based on the embedded semantics of each conversation text obtained in the target subscription content push stage; Based on the information recommendation auxiliary decision-making knowledge, determining the targeted subscription content push tag of the digital cultural and educational interaction log includes: based on the information recommendation auxiliary decision-making knowledge, determining the targeted subscription content push tag of the digital cultural and educational interaction log in the target subscription content push stage.
3. The method according to claim 1, characterized in that Conduct conversation text semantic mining on each collected digital cultural and educational text to obtain conversation text embedded semantics corresponding to each digital cultural and educational text, including: For each of the digital cultural and educational texts, the following processing is performed: For a digital cultural and educational text, performing text word vector mining on the digital cultural and educational text through a deep residual model node to obtain an original text word vector relationship network of the digital cultural and educational text; Enter the original text word vector relationship network into several cascaded vector relationship network processing nodes to obtain the final text word vector relationship network of the last vector relationship network processing node, and obtain the embedded semantics of the conversation text based on the final text word vector relationship network; wherein the processing process of each vector relationship network processing node includes upsampling the entered text word vector relationship network to obtain a first transition text word vector relationship network, performing an expansion sliding filtering operation on the first transition text word vector relationship network to obtain a second transition text word vector relationship network, and performing feature focusing-based processing on the second transition text word vector relationship network to obtain a third transition text word vector relationship network, and performing feature optimization on the third transition text word vector relationship network and the entered text word vector relationship network to obtain a final text word vector relationship network; Among them, for a digital cultural and educational text, a text word vector mining is performed on the digital cultural and educational text through a deep residual model node to obtain an original text word vector relationship network of the digital cultural and educational text, including: Acquire dialogue information corresponding to a user emotion description dimension from dialogue information of a plurality of description dimensions corresponding to the one digital cultural and educational text, wherein the plurality of description dimensions include a user emotion description dimension and a word and sentence logic description dimension; Through the deep residual model node, text word vector mining is performed on the dialogue information corresponding to the user emotion description dimension to obtain the original text word vector relationship network.
4. The method according to any one of claims 1 to 3, characterized in that: Based on the obtained embedded semantics of each conversation text, the corresponding user push preference knowledge representation is obtained through any of the following methods: Integrate the embedded semantics of each conversation text to obtain the user push preference knowledge representation; Performing average processing on the embedded semantics of each conversation text to obtain the knowledge representation of the user push preference; Down-sampling processing is performed on the embedded semantics of each conversation text to obtain the knowledge representation of the user push preference.
5. The method according to claim 1, characterized in that When parsing digital cultural and educational interaction logs, based on the session state style identified in each parsing cycle, a description of the user's language habits in the corresponding parsing cycle is obtained, including: For each parsing cycle, the following operations are performed: Determining whether a semantic gap occurs in a parsing cycle based on a difference between a digital cultural and educational text parsed in a parsing cycle and a digital cultural and educational text parsed in at least one past parsing cycle closest to the parsing cycle; Based on the judgment point of view, the one parsing cycle is marked to obtain a corresponding user language habit description; wherein, when the judgment point of view is continuous, the first language habit variable is marked, and when the judgment point of view is semantic fault, the second language habit variable is marked; Wherein, based on the obtained description of each user's language habits, the corresponding conversation interest linear knowledge representation is obtained, including: Arrange the obtained language habit descriptions of each user according to the priority between the corresponding analysis cycles to obtain a user language habit description sequence; Based on preset description variable mapping features, determining keyword mapping vectors corresponding to a plurality of interest keywords from the user language habit description sequence; Based on the keyword mapping vectors of each of the multiple interest keywords, the conversation interest linear knowledge representation is obtained.
6. A big data push matching system, characterized in that: include: Memory, used to store program instructions and data; A processor, coupled to a memory, and configured to execute instructions in the memory to implement the method according to any one of claims 1 to 5.
7. A computer storage medium, characterized in that: Contains instructions, which, when executed on a processor, implement the method according to any one of claims 1 to 5.
