Digital new media data processing method and device and storage medium

Abstract

The invention relates to the technical field of data processing, in particular to a digital new media data processing method and device and a storage medium, and the method comprises the steps: collecting a comment text and user behavior data; preprocessing the comment contents, constructing a word vector matrix and a user behavior index of each comment content, calculating an emotion extreme value based on the word vector matrix of each comment content to determine a comment type, and performing emotion intensity modeling on each user to determine user emotion intensity; performing user portrait fusion based on the user behavior index, the comment type and the user emotion intensity to determine a user type; generating a comment processing strategy based on the user type and the emotion extreme value of the user comment content in the time window; and updating the generation process of the comment processing strategy of the next management period based on the update factor and the adjustment weight. According to the invention, the processing efficiency of the new media data is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a digital new media data processing method, device and storage medium. Background Art

[0002] With the rapid development of new media platforms, user-generated content (UGC) has seen explosive growth. Efficiently processing massive amounts of comment text, accurately identifying user sentiment, and implementing effective content governance have become core challenges for platform operations. Traditional methods rely on rule-based engines or shallow machine learning models to filter comments through keyword matching or simple sentiment classification (such as positive vs. negative). However, such methods struggle to capture the contextual semantics and emotional intensity of text, leading to misjudgments in complex scenarios such as sarcasm and implicit expressions. Furthermore, the use of user behavior data is limited to a single statistical metric (such as interaction frequency) and lacks integrated analysis of short- and long-term behavioral patterns, resulting in a coarse user profile.

[0003] In existing technologies, policy decisions are mostly based on static threshold settings, which are unable to dynamically adapt to changes in public opinion and fluctuations in user emotions. Furthermore, there is a lack of a feedback mechanism for governance effectiveness, which can easily lead to policy failure or user churn in long-term operations. Furthermore, facing the real-time processing and flexible resource allocation issues of large-scale data, traditional architectures often face the dual bottlenecks of computing efficiency and policy flexibility. Therefore, there is an urgent need for an intelligent processing method that integrates deep semantic analysis, dynamic behavior modeling, and closed-loop policy optimization to improve the accuracy, timeliness, and adaptability of content governance, and achieve the coordinated development of platform ecological health and user experience optimization. Summary of the Invention

[0004] The object of the present invention is to provide a digital new media data processing method, device and storage medium to solve at least one of the problems existing in the prior art.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A digital new media data processing method, comprising: Fusion of user profiles based on user behavior index, comment type, and user sentiment intensity to determine user type; Generate comment processing strategies based on user type and sentiment extremes of user comments within a time window; Target users are screened based on the comment processing strategy within the management cycle, and the target user's emotional fluctuations are calculated based on the emotional extremes of each target user's comment content to determine the adjustment weight. The proportion of negative comments and user retention rate within the management cycle are data-coupled and analyzed to determine the update factor. The generation process of the comment processing strategy for the next management cycle is updated based on the update factor and the adjustment weight.

[0006] Optionally, collect comment text and user behavior data, pre-process the comment content, and construct a word vector matrix and user behavior index for each comment content. Calculate the sentiment extreme value based on the word vector matrix of each comment content to determine the comment type, and simultaneously perform emotion intensity modeling for each user to determine the user's emotion intensity.

[0007] Optionally, use Jieba to segment the comments and use the BERT pre-trained model to generate 768-dimensional sentence vectors to generate a word vector matrix; The user behavior index is constructed based on user behavior data. The expression of the user behavior index is Bi=u1×ai1 / ai2+u2×ln(ai1 / a0+1) / ln2, where Bi is the user behavior index of the i-th user, ai1 is the number of interactions of the i-th user in the past 7 days, ai2 is the total number of interactions of the i-th user, a0 is the comment threshold, u1 is the first weight factor, u2 is the second weight factor, and u1+u2=1.

[0008] Optionally, a BERT model is used to calculate the extreme sentiment value based on the word vector matrix of each comment content, and the extreme sentiment value of the comment content is recorded as Sij. If Sij is less than or equal to the first sentiment discriminant factor s1, the comment type of the comment content is determined to be negative. If Sij is greater than the first sentiment discriminant factor s1 and less than the second sentiment discriminant factor s2, the comment type of the comment content is determined to be neutral. If Sij is greater than or equal to the second sentiment discriminant factor s2, the comment type of the comment content is determined to be positive. Where i represents the user number, j represents the comment content number, and Sij is the extreme sentiment value of the jth comment content of the i-th user in the time window; The user emotion intensity is calculated based on the extreme emotional value of each comment of each user in the time window, and the user emotion intensity of the i-th user is recorded as Fi.

[0009] Optionally, a user type index is constructed based on the user behavior index Bi of the i-th user, the user emotion intensity Fi of the i-th user in the time window, and the comment type. The expression of the user type index is Wi=α1×lg(Bi+1)+α2×sif / ni+α3×Fi, where Wi is the user type index of the i-th user, α1 is the behavior index weight, α2 is the comment type weight, α3 is the user emotion weight, and α1+α2+α3=1; The user type index Wi of the i-th user is compared with the type discrimination factors L1 and L2 to determine the user type. When Wi is less than or equal to L1, the user type is judged to be a low-risk user. When Wi is greater than L1 and less than L2, the user type is judged to be a medium-risk user. When Wi is greater than or equal to L2, the user type is judged to be a high-risk user.

[0010] Optionally, if the user type of the i-th user is a high-risk user and Sij is less than or equal to the risk factor y1, then an intelligent soothing reply is made to the comment content. If the emotional extreme value of the user's comment content is less than or equal to y1 and the number of comments m is greater than or equal to the number threshold M, then the user's emotions are relieved through manual intervention; If the user type of the i-th user is a low-risk user and Sij is greater than or equal to the recommendation factor t1, then the comment will be recommended as a top comment; In other cases, user comments will not be processed.

[0011] Optionally, the number Hi of intelligent soothing replies received by each user during the management period is counted. When Hi is greater than H0, the i-th user is taken as the target user. The average of the extreme emotional values of each comment content of the target user in the first time window of the management period is recorded as Hpi1. The average of the extreme emotional values of each comment content of the target user in the last time window of the management period is recorded as Hpi2. The emotional fluctuation of the target user is set to Bi, Bi=Hpi2-Hpi1; The average value of the emotional fluctuations of all target users is recorded as Bp, and Bp is compared with the emotional fluctuation threshold b0 to determine the adjustment weight. If Bp is less than or equal to b0, the adjustment weight is set to β. Otherwise, the adjustment weight is set to {β×[1+(Bp-b0) 1.5 ]}.

[0012] Optionally, the negative review ratio Fm in the management period is compared with the first proportional factor F1 to determine the negative review factor Fy. If Fm is less than or equal to F0, Fy is set to 0; otherwise, Fy is set to exp[3×(Fm-F0)-3]. Compare the user retention rate Lc with the second scaling factor F2 to determine the retention factor Ly. If Lc is less than F2, set Ly = η × (Lc - F2), where η is the adjustment ratio. Otherwise, set Ly to 0. The negative review factor Fy and the retention rate factor Ly are subjected to data coupling analysis to determine the update factor. The expression of the update factor is Tz=Fy+Ly, where Tz is the update factor. The generation process of the comment processing strategy for the next management cycle is updated based on the update factor and the adjustment weight, and the risk factor y1 is updated to yg. The expression of yg is yg=y1×[1+adjustment weight×lg(3Ly+1) / lg4].

[0013] According to another aspect of the present application, a digital new media data processing device is provided, comprising: Collection unit, used to collect comment text and user behavior data; A data processing unit is used to pre-process the comment content and construct a word vector matrix and user behavior index for each comment content, calculate the extreme sentiment value based on the word vector matrix of each comment content to determine the comment type, and simultaneously perform emotion intensity modeling for each user to determine the user's emotion intensity; The portrait fusion unit is used to fuse user portraits based on user behavior index, comment type, and user emotion intensity to determine the user type; A strategy generation unit, configured to generate a comment processing strategy based on the user type and the extreme sentiment value of the user comment content within a time window; The strategy update unit is used to screen target users according to the comment processing strategy within the management cycle, and calculate the target user's emotional fluctuation based on the emotional extreme values of each comment content of the target user to determine the adjustment weight, conduct data coupling analysis on the proportion of negative comments and user retention rate within the management cycle to determine the update factor, and update the generation process of the comment processing strategy for the next management cycle based on the update factor and adjustment weight.

[0014] According to another aspect of the present application, a computer-readable storage medium is provided, characterized in that the computer-readable storage medium stores a computer program, wherein the computer program is used to control the electronic device where the computer-readable storage medium is located to execute the digital new media data processing method during runtime.

[0015] The beneficial effects of the present invention are as follows: Through the full-link fusion design of data collection, analysis and strategy optimization, an efficient intelligent governance framework is built for new media platforms. At the data layer, the integration and deep cleaning of multi-source heterogeneous data have laid a solid foundation for analysis and eliminated the limitations of traditional single-dimensional evaluation; at the algorithm layer, semantic analysis based on pre-trained models combined with multimodal behavioral feature modeling has achieved accurate deconstruction of user emotional tendencies and behavioral patterns, significantly improving the reliability of emotional polarity classification and risk judgment. The dynamic generation mechanism of user portraits breaks through the mechanical nature of static labeling classification through cross-dimensional joint mapping of behavioral characteristics, emotional intensity and comment attributes, making risk identification more forward-looking and scenario-adaptive, improving the real-time and accuracy of public opinion governance, and promoting the enhancement of platform user stickiness and the healthy development of community ecology through the multi-level release of data value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1Schematic diagram of the process of digital new media data processing according to this embodiment.

[0018] Figure 2 Schematic diagram of the flow of the method for processing comment content in this embodiment.

[0019] Figure 3 Schematic diagram of the process of updating the policy in this embodiment.

[0020] Figure 4 Schematic diagram of the structure of the digital new media data processing device of this embodiment. DETAILED DESCRIPTION

[0021] In order to more clearly illustrate the present invention, the present invention is further described below in conjunction with preferred embodiments and accompanying drawings. Similar components in the accompanying drawings are represented by the same reference numerals. It should be understood by those skilled in the art that the following detailed description is illustrative rather than restrictive and should not be used to limit the scope of protection of the present invention.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] Specifically, the artificial intelligence-based new media data processing method, device and storage medium described in this application are applied to the intelligent governance of user-generated content on digital platforms; the user-generated content described in this application includes user comments; the intelligent governance of new media data in this application involves: deconstruction of comment text features, dynamic modeling of emotional intensity, elastic adaptation of intervention strategies and coupling optimization of ecological indicators. Through the multi-level fusion of semantic enhancement and behavioral characteristics, fine-grained grading of emotional tendencies and accurate identification of user risks are achieved. At the same time, based on the closed-loop feedback mechanism of strategy parameters, adaptive matching of governance intensity and public opinion situation is achieved.

[0024] See also Figure 1 , which is a flow chart of the digital new media data processing method of this embodiment, including: Step S101, collect comment text and user behavior data, the comment text includes user ID, comment content and timestamp, the user behavior data includes the number of interactions in the past 7 days and the total number of interactions, the number of interactions includes likes, reposts, replies, etc.

[0025] For example, in this embodiment, it can be obtained through the user account system, and automatically allocated and stored in the database by the background when the user registers or logs in to the new media platform. The comment content can directly extract the text content from the comment form or API interface submitted by the user. The timestamp is automatically recorded by the background at the precise time (such as UTC time) when the user posts a comment. The number of interactions in the past 7 days and the total number of interactions can be calculated by using SQL time window query or log analysis tools (such as ELK); this embodiment does not specifically limit the method of collecting the above data, and those skilled in the art can freely set it according to needs.

[0026] Step S102: pre-process the comment content, and construct a word vector matrix and user behavior index for each comment content. Calculate the emotional extreme value based on the word vector matrix of each comment content to determine the comment type. Simultaneously, perform emotional intensity modeling on each user to determine the user's emotional intensity.

[0027] See also Figure 2 As shown, the method for processing the comment content includes: Step S201: pre-process the comment content and construct a word vector matrix and user behavior index for each comment content.

[0028] Specifically, step S201 uses Jieba word segmentation to segment the comment content, and uses the BERT pre-training model to generate a 768-dimensional sentence vector to generate a word vector matrix; The user behavior index is constructed based on user behavior data. The expression of the user behavior index is Bi=u1×ai1 / ai2+u2×ln(ai1 / a0+1) / ln2, where Bi is the user behavior index of the i-th user, ai1 is the number of interactions of the i-th user in the past 7 days, ai2 is the total number of interactions of the i-th user, a0 is the comment threshold, u1 is the first weight factor, u2 is the second weight factor, and u1+u2=1.

[0029] Specifically, by integrating semantic analysis with behavioral features, we strengthen our ability to process text data in a structured manner. The application of word segmentation technology and pre-trained models enables deep feature extraction from review texts, effectively capturing implicit semantic information. The dynamic weighting of the behavioral index balances short-term activity and long-term engagement, avoiding biased behavioral assessments and providing multi-dimensional feature support for subsequent analysis.

[0030] For example, in this embodiment, the comment threshold can be set to 50, u1 can be set to 0.7, and u2 can be set to 0.3. In this embodiment, there is no specific limitation on the values of the above data, and those skilled in the art can freely set them according to their needs.

[0031] Illustratively, in this embodiment, when the BERT pre-trained model is used to generate a 768-dimensional sentence vector, [CLS] (classification marker) and [SEP] (separation marker) are respectively inserted at the beginning and end of the text after word segmentation, and the words are mapped to corresponding token IDs through BERT's tokenizer to generate input_ids (token sequence ID), attention_mask (attention mask, distinguishing valid characters from filler characters), and token_type_ids (sentence identifier, all 0s for a single sentence). The above input is sent to the pre-trained BERT model, the last layer of hidden state is output, and the hidden vector corresponding to the [CLS] token (shape: 768 dimensions) is extracted as the sentence vector of the sentence. The [CLS] sentence vectors of all comments are stacked row by row to form a word vector matrix. This embodiment does not specifically limit the method of using the BERT pre-trained model to generate a 768-dimensional sentence vector, and those skilled in the art can freely set it according to their needs.

[0032] Please continue reading Figure 2 As shown, the method for processing the comment content further includes: Step S202 , calculating the extreme sentiment value based on the word vector matrix of each comment content to determine the comment type, and simultaneously performing emotion intensity modeling on each user to determine the user emotion intensity.

[0033] Specifically, step S202 uses the BERT model to calculate the extreme sentiment value based on the word vector matrix of each comment content, and records the extreme sentiment value of the comment content as Sij. If Sij is less than or equal to the first sentiment discriminant factor s1, the comment type of the comment content is determined to be negative. If Sij is greater than the first sentiment discriminant factor s1 and less than the second sentiment discriminant factor s2, the comment type of the comment content is determined to be neutral. If Sij is greater than or equal to the second sentiment discriminant factor s2, the comment type of the comment content is determined to be positive. Where i represents the user number, j represents the comment content number, and Sij is the extreme sentiment value of the jth comment content of the i-th user in the time window; The user emotion intensity is calculated based on the extreme emotional value of each comment of each user in the time window, and the user emotion intensity of the i-th user is recorded as Fi. The expression of Fi is: ; ; Where ni is the number of comments by the i-th user in the time window, and μi is the average of the extreme sentiment values of the i-th user's comments in the time window.

[0034] Specifically, semantically based sentiment extreme value quantification breaks through the crude qualitative limitations of traditional sentiment classification and enables fine-grained grading of review tendencies. Sentiment intensity modeling incorporates temporal sentiment distribution features to reveal the overall trends in user sentiment evolution, enhancing the ability to consistently depict individual emotional states and providing an interpretable quantitative basis for risk prediction.

[0035] For example, in this embodiment, the first emotion discrimination factor s1 can be set to -0.3, and the second emotion discrimination factor s2 can be set to 0.3. In this embodiment, there is no specific limitation on the values of the above data, and those skilled in the art can freely set them according to their needs.

[0036] For example, in this embodiment, when the BERT model is used to calculate the extreme sentiment value based on the word vector matrix of each comment content, a classification layer (such as a fully connected network) is added on top of BERT, and the structure is as follows: [BERT sentence vector (768 dimensions)] → [fully connected layer (768→1)] → [sentiment extreme value Sij (real number)], and a labeled sentiment data set (such as Sij∈[-1,1]) is used for supervised learning to minimize the loss of predicted score and true label, and the sentence vector of each comment in the word vector matrix is input into the trained classification layer, and the corresponding sentiment extreme value Sij is output; this embodiment does not specifically limit the specific method of using the BERT model to calculate the extreme sentiment value based on the word vector matrix of each comment content, and those skilled in the art can freely set it according to their needs.

[0037] Please continue reading Figure 1 As shown, the digital new media data processing method further includes: Step S103: Perform user portrait fusion based on the user behavior index, comment type, and user emotion intensity to determine the user type.

[0038] Specifically, step S104 constructs a user type index based on the user behavior index Bi of the i-th user, the user emotion intensity Fi of the i-th user in the time window, and the comment type. The expression of the user type index is Wi=α1×lg(Bi+1)+α2×sif / ni+α3×Fi, where Wi is the user type index of the i-th user, sif is the number of negative comments of the i-th user in the time window, α1 is the behavior index weight, α2 is the comment type weight, α3 is the user emotion weight, and α1+α2+α3=1; The user type index Wi of the i-th user is compared with the type discrimination factors L1 and L2 to determine the user type. When Wi is less than or equal to L1, the user type is judged to be a low-risk user. When Wi is greater than L1 and less than L2, the user type is judged to be a medium-risk user. When Wi is greater than or equal to L2, the user type is judged to be a high-risk user.

[0039] Specifically, through the combined mapping of multi-dimensional indicators, abstract behavioral data is transformed into concrete user classification criteria. By combining behavioral characteristics, sentiment intensity, and commentary tendencies, this system accurately identifies potentially risky user groups. The hierarchical design of classification results facilitates the adaptation of differentiated management strategies and improves resource allocation efficiency.

[0040] For example, in this implementation, the behavior index weight can be set to 0.6, the comment type weight can be set to 0.2, the user emotion weight can be set to 0.2, L1 can be set to 0.4, and L2 can be set to 0.8; this embodiment does not specifically limit the values of the above data, and those skilled in the art can freely set them according to their needs.

[0041] Please continue reading Figure 1 As shown, the digital new media data processing method further includes: Step S104: generating a comment processing strategy based on the user type and the extreme sentiment value of the user comment content within the time window.

[0042] Specifically, if the user type of the i-th user is a high-risk user and Sij is less than or equal to the risk factor y1, an intelligent soothing reply is made to the comment content. If the emotional extreme value of the user's comment content is less than or equal to y1 and the number of comments m is greater than or equal to the number threshold M, the user's emotions are relieved through manual intervention; If the user type of the i-th user is a low-risk user and Sij is greater than or equal to the recommendation factor t1, then the comment will be recommended as a top comment; In other cases, user comments will not be processed.

[0043] Specifically, automated operational measures are matched to user profiles, enabling rapid response to negative public opinion and intelligent promotion of positive content. A threshold-based decision mechanism balances automated and manual intervention, ensuring efficient handling of high-frequency scenarios while retaining manual flexibility for extreme situations. This differentiated execution of strategies provides users with targeted interactions and enhances platform service awareness.

[0044] For example, in this embodiment, the risk factor y1 can be set to -0.7, the recommendation factor can be set to 0.8, and the quantity threshold M can be set to 5; this embodiment does not specifically limit the values of the above data, and those skilled in the art can freely set them according to needs.

[0045] For example, in this embodiment, when providing intelligent soothing replies to comments, soothing content can be generated in real time based on a pre-trained conversational model (such as GPT-3.5 or T5). For example, given a negative comment and a user type label, the model outputs a friendly response such as "We value your feedback and will improve as soon as possible!" When human intervention is needed to address the user's emotions, a user emotional counseling ticket is automatically generated and pushed to a customer service management system (such as Zendesk or Jira ServiceDesk). The ticket must include the user ID, a summary of recent negative comments, historical soothing records, and a user behavior index. Customer service then contacts the user via in-site message, text message, or phone call to provide a targeted solution.

[0046] Please continue reading Figure 1 As shown, the digital new media data processing method further includes: Step S105, screen the target users based on the comment processing strategy within the management cycle, and calculate the target user's emotional fluctuation based on the emotional extreme value of each comment content of the target user to determine the adjustment weight, conduct data coupling analysis on the proportion of negative comments within the management cycle and the user retention rate to determine the update factor, and update the generation process of the comment processing strategy for the next management cycle based on the update factor and the adjustment weight; the proportion of negative comments within the management cycle is the ratio of the number of comment contents with negative comment type to the total number of comment contents within the management cycle, and the user retention rate is the ratio of the number of users with more than 5 interactions in the last time window within the management cycle to the number of users with more than 5 interactions in the first time window within the management cycle.

[0047] See also Figure 3 As shown, the policy updating method includes: Step S301 : Filter target users based on the comment processing strategy within the management cycle, and calculate the target user's emotional fluctuation based on the emotional extreme values of each comment content of the target user to determine the adjustment weight.

[0048] Specifically, step S107 counts the number Hi of intelligent soothing replies received by each user during the management period. When Hi is greater than H0, the i-th user is taken as the target user, and the average of the extreme emotional values of each comment content of the target user in the first time window of the management period is recorded as Hpi1, and the average of the extreme emotional values of each comment content of the target user in the last time window of the management period is recorded as Hpi2, and the emotional fluctuation of the target user is set to Bi, Bi=Hpi2-Hpi1; The average value of the emotional fluctuations of all target users is recorded as Bp, and Bp is compared with the emotional fluctuation threshold b0 to determine the adjustment weight. If Bp is less than or equal to b0, the adjustment weight is set to β. Otherwise, the adjustment weight is set to {β×[1+(Bp-b0) 1.5 ]}.

[0049] Specifically, it identifies high-frequency intervention user groups, focusing on core governance targets to optimize resource allocation. Sentiment analysis monitors user sentiment, quantifies the actual effectiveness of policy interventions, and supports dynamic weighting. A flexible parameter setting mechanism enhances policy adaptability, achieving a precise match between governance intensity and user status.

[0050] For example, in this embodiment, the emotional fluctuation threshold can be set to 0.3, and the baseline adjustment weight β can be set to 0.2; in this embodiment, there is no specific limitation on the values of the above data, and those skilled in the art can freely set them according to needs.

[0051] See also Figure 3 As shown, the strategy updating method further includes: Step S302: Perform data coupling analysis on the proportion of negative comments and user retention rate within the management cycle to determine the update factor, and update the generation process of the comment processing strategy for the next management cycle based on the update factor and the adjustment weight.

[0052] Specifically, step S106 compares the negative review ratio Fm within the management period with the first proportional factor F1 to determine the negative review factor Fy. If Fm is less than or equal to F0, Fy is set to 0; otherwise, Fy is set to exp[3×(Fm-F0)-3]. Compare the user retention rate Lc with the second scaling factor F2 to determine the retention factor Ly. If Lc is less than F2, set Ly = η × (Lc - F2), where η is the adjustment ratio. Otherwise, set Ly to 0. The negative review factor Fy and the retention rate factor Ly are subjected to data coupling analysis to determine the update factor. The expression of the update factor is Tz=Fy+Ly, where Tz is the update factor. The generation process of the comment processing strategy for the next management cycle is updated based on the update factor and the adjustment weight, and the risk factor y1 is updated to yg. The expression of yg is yg=y1×[1+adjustment weight×lg(3Ly+1) / lg4].

[0053] Specifically, multi-dimensional verification of the proportion of negative content and user engagement data reveals the profound impact of the strategy on the platform ecosystem. Based on the targeted integration of negative feedback intensity and retention levels, we drive targeted optimization of strategy parameters and establish a response chain from feedback results to decision adjustments. The cyclical iterative design of model factors forms a self-evolving strategy mechanism, continuously improving the system's adaptability and foresight in complex scenarios.

[0054] For example, in this embodiment, the first scale factor can be set to 0.05, the second scale factor can be set to 0.7, and the adjustment ratio can be set to 0.5. In this embodiment, the values of the above data are not specifically limited, and those skilled in the art can freely set them according to their needs. For example, in this embodiment, the time window can be set to 5 days and the management period can be set to 40 days; in this embodiment, there is no specific limitation on the setting of the time window and the management period, and those skilled in the art can freely set them according to their needs.

[0055] See also Figure 4 As shown, the present application also provides a digital new media data processing device, comprising: Collection unit 501, used to collect comment text and user behavior data; Data processing unit 502 is used to pre-process the comment content and construct a word vector matrix and user behavior index for each comment content, calculate the extreme sentiment value based on the word vector matrix of each comment content to determine the comment type, and simultaneously perform emotion intensity modeling for each user to determine the user's emotion intensity; A portrait fusion unit 503 is used to fuse user portraits based on the user behavior index, comment type, and user emotion intensity to determine the user type; A strategy generating unit 504 is configured to generate a comment processing strategy based on the user type and the extreme sentiment value of the user comment content within the time window; The strategy update unit 505 is used to screen target users according to the comment processing strategy within the management cycle, and calculate the target user's emotional fluctuation based on the emotional extreme values of each comment content of the target user to determine the adjustment weight, conduct data coupling analysis on the proportion of negative comments and user retention rate within the management cycle to determine the update factor, and update the generation process of the comment processing strategy for the next management cycle based on the update factor and the adjustment weight.

[0056] The present application also provides a computer-readable storage medium, which is a tangible physical storage medium that can store the above-mentioned computer program and various types of data used in the program; the physical storage medium includes but is not limited to existing physical storage media such as random access memory, read-only memory, optical disk, hard disk, or a combination of media.

[0057] Those skilled in the art will appreciate that all or some of the steps and systems in the method disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include computer storage media (or non-transitory media) and communication media (or temporary media). As known to those skilled in the art, the term computer storage media is included in any method or technology for storing information (such as a computer-readable program, a data structure, a program module, or other data) and is volatile and non-volatile, removable, and non-removable. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (DVD), or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage, or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media typically embodies computer-readable programs, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0058] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above implementation. Those skilled in the art can also make various equivalent modifications or substitutions under the shared conditions that do not violate the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A digital new media data processing method, characterized in that: include: Fusion of user profiles based on user behavior index, comment type, and user sentiment intensity to determine user type; Generate comment processing strategies based on user type and sentiment extremes of user comments within a time window; Target users are screened based on the comment processing strategy within the management cycle, and the target user's emotional fluctuations are calculated based on the emotional extremes of each target user's comment content to determine the adjustment weight. The proportion of negative comments and user retention rate within the management cycle are data-coupled and analyzed to determine the update factor. The generation process of the comment processing strategy for the next management cycle is updated based on the update factor and the adjustment weight.

2. The digital new media data processing method according to claim 1, characterized in that: Also includes: Collect comment text and user behavior data, pre-process the comment content, and construct the word vector matrix and user behavior index of each comment content. Calculate the sentiment extreme value based on the word vector matrix of each comment content to determine the comment type, and model the emotion intensity of each user to determine the user's emotion intensity.

3. The digital new media data processing method according to claim 2, characterized in that: Use Jieba word segmentation to segment the comments, and use the BERT pre-trained model to generate 768-dimensional sentence vectors to generate word vector matrices; The user behavior index is constructed based on user behavior data. The expression of the user behavior index is Bi=u1×ai1 / ai2+u2×ln(ai1 / a0+1) / ln2, where Bi is the user behavior index of the i-th user, ai1 is the number of interactions of the i-th user in the past 7 days, ai2 is the total number of interactions of the i-th user, a0 is the comment threshold, u1 is the first weight factor, u2 is the second weight factor, and u1+u2=1.

4. The digital new media data processing method according to claim 3, characterized in that: Use the BERT model to calculate the extreme sentiment value based on the word vector matrix of each comment content, and record the extreme sentiment value of the comment content as Sij. If Sij is less than or equal to the first sentiment discriminant factor s1, the comment type of the comment content is judged to be negative. If Sij is greater than the first sentiment discriminant factor s1 and less than the second sentiment discriminant factor s2, the comment type of the comment content is judged to be neutral. If Sij is greater than or equal to the second sentiment discriminant factor s2, the comment type of the comment content is judged to be positive. Where i represents the user number, j represents the comment content number, and Sij is the extreme sentiment value of the jth comment content of the i-th user in the time window; The user emotion intensity is calculated based on the extreme emotional value of each comment of each user in the time window, and the user emotion intensity of the i-th user is recorded as Fi.

5. The digital new media data processing method according to claim 4, characterized in that: The user type index is constructed based on the user behavior index Bi of the i-th user, the user emotion intensity Fi of the i-th user in the time window, and the comment type. The expression of the user type index is Wi = α1×lg(Bi+1)+α2×sif / ni+α3×Fi, where Wi is the user type index of the i-th user, α1 is the behavior index weight, α2 is the comment type weight, and α3 is the user emotion weight. α1+α2+α3=1; The user type index Wi of the i-th user is compared with the type discrimination factors L1 and L2 to determine the user type. When Wi is less than or equal to L1, the user type is judged to be a low-risk user. When Wi is greater than L1 and less than L2, the user type is judged to be a medium-risk user. When Wi is greater than or equal to L2, the user type is judged to be a high-risk user.

6. The digital new media data processing method according to claim 5, characterized in that: If the user type of the i-th user is a high-risk user and Sij is less than or equal to the risk factor y1, then an intelligent soothing reply is made to the comment content. If the emotional extreme value of the user's comment content is less than or equal to y1 and the number of comments m is greater than or equal to the number threshold M, then the user's emotions are relieved through manual intervention; If the user type of the i-th user is a low-risk user and Sij is greater than or equal to the recommendation factor t1, then the comment will be recommended as a top comment; In other cases, user comments will not be processed.

7. The digital new media data processing method according to claim 6, characterized in that: Count the number of intelligent soothing replies Hi received by each user during the management cycle. When Hi is greater than H0, the i-th user is taken as the target user. The average emotional extreme values of each comment content of the target user in the first time window of the management cycle is recorded as Hpi1. The average emotional extreme values of each comment content of the target user in the last time window of the management cycle is recorded as Hpi2. The emotional fluctuation of the target user is set to Bi, Bi=Hpi2-Hpi1; The average value of the emotional fluctuations of all target users is recorded as Bp, and Bp is compared with the emotional fluctuation threshold b0 to determine the adjustment weight. If Bp is less than or equal to b0, the adjustment weight is set to β. Otherwise, the adjustment weight is set to {β×[1+(Bp-b0) 1.5 ]}.

8. The digital new media data processing method according to claim 7, characterized in that: Compare the negative review ratio Fm within the management period with the first proportional factor F1 to determine the negative review factor Fy. If Fm is less than or equal to F0, set Fy to 0; otherwise, set Fy = exp[3×(Fm-F0)-3]. Compare the user retention rate Lc with the second scaling factor F2 to determine the retention factor Ly. If Lc is less than F2, set Ly = η × (Lc - F2), where η is the adjustment ratio. Otherwise, set Ly to 0. The negative review factor Fy and the retention rate factor Ly are subjected to data coupling analysis to determine the update factor. The expression of the update factor is Tz=Fy+Ly, where Tz is the update factor. The generation process of the comment processing strategy for the next management cycle is updated based on the update factor and the adjustment weight, and the risk factor y1 is updated to yg. The expression of yg is yg=y1×[1+adjustment weight×lg(3Ly+1) / lg4].

9. A digital new media data processing device, characterized in that: include: Collection unit, used to collect comment text and user behavior data; A data processing unit is used to pre-process the comment content and construct a word vector matrix and user behavior index for each comment content, calculate the extreme sentiment value based on the word vector matrix of each comment content to determine the comment type, and simultaneously perform emotion intensity modeling for each user to determine the user's emotion intensity; The portrait fusion unit is used to fuse user portraits based on user behavior index, comment type, and user emotion intensity to determine the user type; A strategy generation unit, configured to generate a comment processing strategy based on the user type and the extreme sentiment value of the user comment content within a time window; The strategy update unit is used to screen target users according to the comment processing strategy within the management cycle, and calculate the target user's emotional fluctuation based on the emotional extreme values of each comment content of the target user to determine the adjustment weight, conduct data coupling analysis on the proportion of negative comments and user retention rate within the management cycle to determine the update factor, and update the generation process of the comment processing strategy for the next management cycle based on the update factor and adjustment weight.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program is used to control the electronic device where the computer-readable storage medium is located to execute the digital new media data processing method according to any one of claims 1 to 8 during operation.

Images