A dynamic analysis method and computer system for consumer big data

Abstract

The embodiments of the present invention disclose a dynamic analysis method and computer system for consumer big data. The method can comprehensively collect various types of information during the consumption process by obtaining a consumer feedback text collection containing multiple consumer interaction records in a target consumption scenario. Dynamic semantic analysis of the consumer feedback text collection can accurately extract the semantic and emotional tendency feature collection of each consumer interaction record to deeply explore the connotation of user feedback. Based on the dynamic correlation mapping processing of semantic and emotional tendency features, a service optimization feature collection can be generated in a targeted manner, providing strong support for accurately positioning the direction of service improvement. Then, a dynamic adjustment strategy for consumer services is generated, and the interaction process of the consumer service system is optimized, thereby effectively improving the adaptability and flexibility of the consumer service system.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of big data analysis, and specifically to a dynamic analysis method and computer system for consumer big data. Background Art

[0002] In the field of consumer big data analysis, with the rapid development of information technology, massive amounts of consumer data are continuously generated. Analyzing this data can help better understand consumer needs and improve service quality. Existing consumer big data analysis technologies can collect and organize consumer data to a certain extent, such as simply counting the frequency of consumer behavior. However, existing technologies have significant shortcomings. First, they lack in-depth processing of the textual information in consumer feedback, remaining limited to superficial keyword extraction and failing to fully understand the true intent and emotional tendencies behind user feedback statements. Second, they struggle to effectively correlate the different features in consumer feedback, making it impossible to accurately identify the key areas for service improvement. Consequently, existing technologies are unable to dynamically generate effective service adjustment strategies based on consumer feedback, and they also lack specificity and flexibility in optimizing the interactive processes of consumer service systems. Summary of the Invention

[0003] The embodiments of the present invention provide a dynamic analysis method and computer system for consumer big data, which are used to optimize the interactive process of a consumer service system, thereby effectively improving the adaptability and flexibility of the consumer service system.

[0004] In a first aspect, an embodiment of the present invention provides a dynamic analysis method for consumer big data, which is applied to a computer system. The method includes: obtaining a consumer feedback text set in a target consumption scenario, the consumer feedback text set including multiple consumer interaction records, each consumer interaction record consisting of at least one user feedback statement and a corresponding consumer service identifier; performing dynamic semantic analysis processing on the consumer feedback text set to obtain a semantic feature set and an emotional tendency feature set for each consumer interaction record; generating a service optimization feature set for the consumer interaction record based on dynamic association mapping processing between the semantic feature set and the emotional tendency feature set; generating a consumer service dynamic adjustment strategy based on the service optimization feature set, and using the consumer service dynamic adjustment strategy to optimize the interaction process of the consumer service system corresponding to the target consumption scenario.

[0005] In a second aspect, an embodiment of the present invention provides a computer system, including:

[0006] processor;

[0007] a storage device having a computer program stored thereon,

[0008] When the computer program is executed by the processor, the processor implements any of the dynamic analysis methods for consumption big data.

[0009] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the dynamic analysis method of consumption big data are implemented.

[0010] Thus, the embodiments of the present invention have the following beneficial effects: by obtaining a consumer feedback text set containing multiple consumer interaction records in the target consumption scenario, various types of information in the consumption process can be comprehensively collected. Dynamic semantic analysis of the consumer feedback text set can accurately extract the semantic and emotional tendency feature set of each consumer interaction record to deeply explore the connotation of user feedback. Based on the dynamic correlation mapping processing of semantic and emotional tendency features, a service optimization feature set can be generated in a targeted manner, providing strong support for accurately locating the direction of service improvement. Then, a dynamic adjustment strategy for consumer services is generated, and the interaction process of the consumer service system is optimized, thereby effectively improving the adaptability and flexibility of the consumer service system. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a flow chart of a method for dynamic analysis of consumption big data provided by an embodiment of the present invention.

[0012] Figure 2 A schematic diagram of the basic structure of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0014] See also Figure 1 As shown in FIG, this figure is a flow chart of a method for dynamic analysis of consumption big data provided by an embodiment of the present invention, which can be applied in a computer system. Figure 1 As shown, the method may include steps S110 to S140.

[0015] Step S110: obtaining a consumption feedback text set in a target consumption scenario, wherein the consumption feedback text set includes a plurality of consumption interaction records, and each consumption interaction record is composed of at least one user feedback statement and a corresponding consumption service identifier.

[0016] In this embodiment, taking the e-commerce consumption scenario as an example, the target consumption scenario focuses on the product purchase process of the e-commerce platform system. The process of obtaining the consumer feedback text set is as follows: the e-commerce platform system sets a corresponding data collection module to collect in real time the public feedback information submitted by users after completing the product purchase interaction process. For example, after purchasing an electronic product, user A submits the feedback statement "The payment page jump is not smooth during the purchase process, and the waiting time is long", and the corresponding consumer service identifier is "Payment page interaction guidance"; after purchasing clothing, user B submits feedback "The drop-down menu for selecting size is inconvenient to operate", and its consumer service identifier is "Size selection interface interaction". In this way, a consumer feedback text set containing many similar consumer interaction records is continuously accumulated.

[0017] Step S120: performing dynamic semantic analysis on the consumer feedback text set to obtain a semantic feature set and a sentiment tendency feature set of each consumer interaction record.

[0018] In this step, the e-commerce platform system's data processing module conducts a comprehensive analysis of the collected consumer feedback text. Taking user A's feedback statement, "The payment page during the purchase process was choppy and the wait time was long," as an example, the system aims to uncover the semantic connotations and underlying sentiment of this statement. This allows for a deeper understanding of the core issues raised by the user feedback and their emotional attitudes toward the relevant interaction process, providing guidance for subsequent optimization efforts.

[0019] In an alternative implementation, the dynamic semantic analysis processing is performed on the consumer feedback text set to obtain a semantic feature set and a sentiment tendency feature set of each consumer interaction record, including:

[0020] Step S121: performing text unit division processing on the user feedback sentences in the consumption interaction record to obtain a plurality of semantically associated text unit sequences.

[0021] The system then divides user A's feedback statement according to semantic logic. For example, it can be divided into text units such as "purchase process," "payment page," "unsmooth transition," and "long wait time." These text units are closely semantically linked. "Purchase process" clarifies the broad scope of the feedback, "payment page" further refines it to the specific interactive interface, and "unsmooth transition" and "long wait time" specifically describe the issues encountered on that page. This division creates a semantically coherent sequence of text units, providing a clear structure for subsequent semantic encoding processing.

[0022] Step S122: calling a pre-trained semantic encoding model to perform context encoding processing on the text unit sequence to generate context semantic features of the user feedback sentence; wherein the context semantic features include the semantic association and weight distribution between different text units in the text unit sequence.

[0023] The pre-trained semantic coding model of this embodiment is trained in advance by the e-commerce platform system based on a large amount of text data. For the divided text unit sequence, the semantic coding model will first analyze the meaning of each text unit. For example, "payment page" is related to many page-related concepts in the e-commerce platform, and the semantic coding model understands its semantics through learned knowledge. Next, the semantic coding model calculates the semantic correlation between different text units. For example, there is a direct causal relationship between "payment page" and "unsmooth jump", which means that there is a problem with the jump on the payment page. At the same time, the semantic coding model assigns weights according to the importance of the text unit in the sentence. "Unsmooth jump" and "long waiting time" are more critical in describing the problem and have relatively high weights, while "purchase process" as a background description has a relatively low weight. Through the above processing, contextual semantic features containing semantic correlation and weight distribution are generated.

[0024] In a preferred implementation, calling a pre-trained semantic encoding model to perform context encoding processing on the text unit sequence to generate contextual semantic features of the user feedback sentence includes:

[0025] Step S1221: performing vectorization conversion processing on each text unit in the text unit sequence to generate an initial text unit vector.

[0026] Optionally, the system maps each text unit into a vector space, generating a corresponding vector representation for each text unit. For example, "purchase process" might be converted into a vector of corresponding dimensions, whose numerical distribution represents the position and characteristics of "purchase process" in the semantic space. Each text unit undergoes this transformation to form an initial text unit vector, which serves as the basic data form for subsequent processing, facilitating model calculations and analysis.

[0027] Step S1222: constructing a position coding feature of the initial text unit vector based on the position correlation between adjacent text units in the text unit sequence.

[0028] Given the importance of the position of text units within a sequence, the system analyzes the positional relationships between adjacent text units. For example, the "payment page" appears after the "purchase process," and this order reflects a certain logical relationship. Based on this positional correlation, the system constructs a position encoding feature for the initial text unit vector. This feature incorporates position information into the vector, allowing the model to consider the order of text units when processing the vector, better capturing semantic information.

[0029] Step S1223: superimpose the initial text unit vector and the position coding feature to obtain a coding input vector.

[0030] In this step, the generated initial text unit vector and position encoding feature can be superimposed. For example, the initial text unit vector of the "payment page" and its corresponding position encoding feature vector are added according to certain rules to obtain a new vector, namely the encoding input vector. This encoding input vector contains both the semantic information of the text unit itself and the position information, providing more comprehensive input data for subsequent semantic attention calculations.

[0031] Step S1224: calling the semantic encoding model to perform multi-layer semantic attention calculation on the encoded input vector to generate contextual semantic features containing global semantic relevance.

[0032] The semantic encoding model performs semantic attention calculations on the encoded input vector at multiple levels. At each level, the model focuses on different parts of the vector to capture more comprehensive semantic information. For example, in the first level of attention calculation, the model might focus on the portion of the vector related to the "payment page" and analyze its local association with other text unit vectors. As the number of levels increases, the model gradually considers the entire vector sequence and calculates global semantic association. Through multi-level calculations, a contextual semantic feature containing global semantic association is ultimately generated, which can more accurately reflect the overall semantic information of the user feedback sentence.

[0033] In an exemplary embodiment, calling the semantic encoding model to perform multi-layer semantic attention calculation on the encoded input vector to generate contextual semantic features containing global semantic relevance includes:

[0034] Step S12240: In each semantic attention layer of the semantic coding model, multi-head attention calculation is performed on the coding input vector to generate intermediate attention features; residual connection processing is performed on the intermediate attention features and the coding input vector to obtain residual features; layer normalization processing is performed on the residual features to obtain normalized features; the normalized features are input into a feedforward neural network for nonlinear transformation to obtain output features of the current semantic attention layer; multi-layer semantic attention calculation is iteratively performed until a preset number of layers is reached, and the output features of the last layer are used as the contextual semantic features.

[0035] At each semantic attention layer, the model performs multi-head attention calculations, focusing on the encoded input vector from different perspectives. For example, one head may focus on the key words in the vector, while another head focuses on the overall structural information of the vector, generating intermediate attention features in this way. The intermediate attention features are then residually connected with the encoded input vector, retaining some important information in the original vector to obtain residual features. The residual features are then layer-normalized to make the data distribution more stable, resulting in normalized features. The normalized features are input into a feedforward neural network for nonlinear transformation, increasing the model's expressive power, and obtaining the output features of the current layer. This process is repeated until the preset number of layers is reached. The output features of the last layer can be used as contextual semantic features that contain global semantic relevance and can accurately reflect the semantic overview of the user feedback sentence.

[0036] Step S123: performing sentiment polarity recognition processing on the user feedback sentence to generate sentiment tendency features of the user feedback sentence; the sentiment tendency features include at least one of the following: sentiment polarity intensity, sentiment category distribution, and sentiment fluctuation trend of the user feedback sentence.

[0037] Optionally, the system performs sentiment analysis on user feedback to determine the user's emotional attitude. For user A's feedback, sentiment analysis techniques are used to determine its emotional characteristics. For example, the sentiment polarity of the statement is determined to be strongly negative because the user explicitly identified issues with the payment page; the sentiment category is characterized by dissatisfaction with the interaction process; and the sentiment fluctuation trend is relatively stable due to the specific nature of the feedback. These emotional characteristics can help e-commerce platforms understand users' feelings about the interaction process and provide guidance for improvement.

[0038] In an alternative embodiment, performing sentiment polarity recognition processing on the user feedback sentence to generate the sentiment tendency feature of the user feedback sentence includes:

[0039] Step S1231: performing emotional keyword extraction processing on the user feedback sentence to obtain an emotional keyword set and corresponding emotional weights.

[0040] The system extracts emotional keywords from user feedback. For User A's feedback, words like "unfluent" and "long" reflect negative emotions and are extracted as emotional keywords. These words are also weighted based on their importance in expressing emotion. "Unfluent" is more critical to expressing negative emotions and has a relatively higher weight. This results in a set of emotional keywords and their corresponding emotional weights, providing a foundation for subsequent emotional polarity labeling.

[0041] Step S1232: Based on the emotion knowledge base and the emotion weight, the emotion keyword set is subjected to emotion polarity labeling processing to generate an emotion polarity distribution vector.

[0042] Optionally, a pre-built sentiment knowledge base can be used, combined with sentiment weights, to assign polarity labels to sentiment keywords. The sentiment knowledge base stores sentiment polarity information for various words; for example, "unfluent" is labeled as negative. Based on the sentiment weights, the polarity of each keyword is quantified to generate a vector. Each dimension of the vector represents the degree of polarity, such as positive, negative, or neutral. This vector can be used as a sentiment polarity distribution vector, visually displaying the sentiment polarity distribution of user feedback sentences.

[0043] Step S1233: performing temporal sentiment analysis on the user feedback statement to generate sentiment fluctuation sequence features; wherein the sentiment fluctuation sequence features reflect the temporal variation pattern of the sentiment polarity in the user feedback statement.

[0044] If user feedback statements represent interactions recorded over a longer period of time, the system performs time-series sentiment analysis. For example, for a series of feedback statements about the purchase process, the system analyzes sentiment changes chronologically. By analyzing sentiment keywords and sentiment polarity at different time points, it generates sentiment fluctuation sequence features. This feature can help e-commerce platforms understand the changing trends of user sentiment over time, such as whether negative sentiment increases over time.

[0045] In a preferred embodiment, performing temporal sentiment analysis on the user feedback sentences to generate sentiment fluctuation sequence features includes:

[0046] Step S12330: Divide the user feedback statement into multiple sentiment analysis windows in chronological order; in each sentiment analysis window, call the sentiment classification model to predict the sentiment polarity of the text content in the current sentiment analysis window to generate a window sentiment label; perform differential calculation on the sentiment labels of adjacent windows based on the sliding window mechanism to generate a sentiment label change rate sequence; perform trend fitting processing on the sentiment label change rate sequence to generate a sentiment fluctuation sequence feature that reflects the direction and intensity of sentiment fluctuation.

[0047] Optionally, the system divides user feedback statements into multiple windows in chronological order, for example, each interaction as a window. Within each window, a sentiment classification model is used to predict sentiment polarity, resulting in a window sentiment label, such as "negative" or "neutral." A sliding window mechanism is then used to calculate the differences in sentiment labels between adjacent windows, yielding a sequence of sentiment label change rates. Finally, this sequence is subjected to trend fitting, for example using polynomial fitting, to obtain sentiment fluctuation sequence features that reflect the direction and intensity of sentiment fluctuations, clearly demonstrating how user sentiment changes over time.

[0048] Step S1234: performing joint encoding processing on the emotion polarity distribution vector and the emotion fluctuation sequence feature to generate the emotion tendency feature.

[0049] In this embodiment, the emotion polarity distribution vector and the emotion fluctuation sequence feature are jointly encoded. For example, the two vectors are concatenated to form a new vector, which can be used as a complete emotion tendency feature. This new vector combines the emotion polarity distribution and emotion fluctuation information to more comprehensively describe the emotional tendency of the user feedback sentence, providing rich data support for subsequent analysis and processing.

[0050] Step S124: performing feature fusion processing on the contextual semantic features and the emotional tendency features to obtain a semantic feature set and an emotional tendency feature set of the consumption interaction record.

[0051] Specifically, the system integrates the previously generated contextual semantic features and sentiment features. For example, the contextual semantic feature vector and sentiment feature vector are concatenated or weighted summed according to certain rules to form a new feature set. This set contains both the semantic and sentiment information of the user feedback sentence, forming the semantic and sentiment feature sets of the consumer interaction record, providing a comprehensive data foundation for the subsequent generation of service optimization feature sets.

[0052] Step S130: generating a service optimization feature set of the consumption interaction record based on a dynamic association mapping process between the semantic feature set and the emotional tendency feature set.

[0053] The e-commerce platform system's data processing module deeply analyzes the relationship between semantic feature sets and sentiment feature sets to generate a service optimization feature set. For example, by combining semantic features and sentiment features generated from user A's feedback, the system can uncover the correlation between the payment page interaction process and the user's negative emotions, thereby identifying areas for optimization and generating the corresponding service optimization feature set.

[0054] As a preferred embodiment, the generating of the service optimization feature set of the consumption interaction record based on the dynamic association mapping process between the semantic feature set and the emotional tendency feature set includes:

[0055] Step S131: Calculating importance weights of the contextual semantic features in the semantic feature set to generate an importance weight distribution.

[0056] The system analyzes the importance of each component of the contextual semantic features. For example, in the contextual semantic features reported by User A, "unsmooth transitions" and "long wait times" are more critical in describing issues with the payment page and are assigned higher importance weights. Meanwhile, "purchase process," as background information, has a relatively lower weight. This calculation generates an importance weight distribution, clarifying the importance of each semantic component within the overall picture.

[0057] Step S132: performing dynamic weighted fusion processing on the emotional tendency feature set according to the importance weight distribution to generate a joint emotional semantic feature.

[0058] As you can understand, based on the previously calculated importance weights, the sentiment feature set is weighted and fused. For example, if the weight for "unsmooth transitions" is high, then the associated negative sentiment polarity will account for a larger proportion in the fusion. This dynamic weighted fusion generates a joint sentiment semantic feature that integrates semantic and sentiment information, more accurately reflecting the relationship between user feedback and service issues.

[0059] Step S133: performing association mapping processing on the joint emotional semantic feature and the consumption service identifier to generate a service association feature; wherein the service association feature reflects the matching degree and optimization priority between the user feedback statement and the consumption service identifier.

[0060] Furthermore, the combined sentiment semantic feature is associated with the consumer service identifier, "Payment Page Interaction Guidance." The system calculates the degree of match between them. For example, by analyzing the correlation between semantics and sentiment and the service identifier, it can determine that the feedback has a high match with the payment page interaction guidance. Furthermore, due to the user's negative sentiment, this service also has a high optimization priority. Based on this association, a service-related feature is generated, providing a clear direction for subsequent service optimization.

[0061] In an optional embodiment, the step of performing an associative mapping process on the combined emotional semantic feature and the consumption service identifier to generate a service-associated feature includes:

[0062] Step S1330: Obtain a service attribute feature set corresponding to the consumer service identifier; the service attribute feature set includes at least one of the following: service response timeliness feature, service coverage feature and service resource allocation feature; calculate the multidimensional similarity matrix between the joint emotional semantic feature and the service attribute feature set; perform feature screening on the multidimensional similarity matrix, and retain the associated dimensions that meet the preset similarity threshold; construct the service association feature between the user feedback statement and the consumer service identifier based on the retained associated dimensions.

[0063] Specifically, the system first obtains a set of service attribute features corresponding to "payment page interaction guidance," such as service response timeliness (e.g., payment page loading speed). It then calculates a multidimensional similarity matrix between the joint sentiment semantic features and these service attribute features, with each element in this matrix representing the similarity between the two features. Next, it selects associated dimensions that meet a preset similarity threshold, for example, retaining only those dimensions with a high correlation between sentiment and service attributes. Finally, based on these retained dimensions, it constructs service-related features to more accurately reflect the relationship between user feedback and the service.

[0064] Step S134: performing dimensionality reduction and compression processing on the service-related features to generate the service optimization feature set.

[0065] Because service-related features can be high-dimensional and difficult to process, the system reduces and compresses them. For example, using methods like principal component analysis, high-dimensional service-related features can be converted into a low-dimensional feature set while retaining the key information. This reduced-dimensional set can then be used as a service optimization feature set, providing concise and critical data support for generating dynamic adjustment strategies for consumer services.

[0066] Step S140: generating a consumption service dynamic adjustment strategy based on the service optimization feature set, and utilizing the consumption service dynamic adjustment strategy to optimize the interaction process of the consumption service system corresponding to the target consumption scenario.

[0067] Optionally, the e-commerce platform can formulate a dynamic adjustment strategy for consumer services based on the generated service optimization feature set and apply this strategy to optimize the interaction process of the consumer service system. For example, to address issues with the payment page, a corresponding optimization strategy can be formulated to improve the user experience.

[0068] In a possible embodiment, generating a consumption service dynamic adjustment strategy according to the service optimization feature set includes:

[0069] Step S141: performing cluster analysis on the service optimization feature set to generate multiple service optimization categories and corresponding category center features.

[0070] Optionally, the system performs cluster analysis on the service optimization feature set, for example using the K-means clustering algorithm. Similar service optimization features are grouped together to form different service optimization categories. Each category has a category-centric feature. For example, a category might focus on payment page performance optimization, and its category-centric feature represents the typical characteristics of service optimization in that category.

[0071] Step S142: performing a matching search based on the category center feature and the historical optimization strategy library to determine an initial service adjustment strategy set.

[0072] Optionally, the generated category center features are compared with a historical optimization strategy library. This library stores past optimization strategies for various service issues. For example, for the payment page performance optimization category, strategies that have successfully solved similar issues are found in the library to determine the initial set of service adjustment strategies, providing a reference for subsequent adjustments.

[0073] Step S143: Prioritizing the initial service adjustment policy set according to the real-time consumption service load status to generate a policy execution sequence.

[0074] Consider the real-time load status of consumer services, such as the number of visits to the current payment page. If the number of visits is high, then policies that can quickly improve the response speed of the payment page will be prioritized. This method sorts the initial set of service adjustment policies and generates a policy execution sequence, ensuring that the most effective policies for the current situation are executed first.

[0075] Step S144: taking the subset of policies that meet the real-time load constraint in the policy execution sequence as the consumer service dynamic adjustment policy;

[0076] Filter out a subset of policies from the policy execution sequence that meet real-time load constraints. For example, if the current server resources are limited, some resource-intensive policies may not be executed. In this case, only the policies that can be executed under the existing resource conditions are selected. This subset can be used as a consumer service to dynamically adjust the policy.

[0077] In another possible embodiment, the utilizing the consumption service dynamic adjustment strategy to optimize the interaction process of the consumption service system corresponding to the target consumption scenario includes:

[0078] Step S145: obtaining a service adjustment parameter mapping table in the consumption service dynamic adjustment strategy, wherein the service adjustment parameter mapping table includes service resource configuration parameters and interactive response rules corresponding to a plurality of service optimization categories.

[0079] Optionally, a service adjustment parameter mapping table is obtained from the consumer service dynamic adjustment policy. For example, for the payment page performance optimization service category, the mapping table will record relevant service resource configuration parameters, such as the number of server threads allocated to payment page processing and computing resource quotas. It also includes interaction response rules, such as the maximum response time for the payment page. These parameters and rules provide specific guidance for optimizing the interaction process of the consumer service system.

[0080] Step S146: Dynamically allocate resources to the real-time service load indicators of the consumer service system according to the service resource configuration parameters to generate resource allocation optimization results; wherein the dynamic resource allocation processing includes adjusting at least one of the number of service nodes, computing resource quotas and service thread priorities.

[0081] In this embodiment, dynamic resource allocation is performed based on the service resource configuration parameters in the service adjustment parameter mapping table, combined with the real-time service load indicators of the consumer service system. If the real-time service load of the current payment page shows a surge in visits, the number of service nodes can be increased based on the parameters, for example, from the original 5 server nodes to 8 to share the processing pressure; or the computing resource quota can be adjusted to allocate more CPU and memory resources to payment page processing; or the priority of the relevant service thread can be increased to ensure that payment page requests are processed first. Through these operations, resource allocation optimization results are generated, improving the system's ability to cope with the current load and ensuring a smooth interaction process.

[0082] Step S147: Based on the interactive response rules, the interactive process control parameters of the consumer service system are updated to generate an updated interactive process control strategy; the interactive process control parameters include at least one of the following: user request routing strategy, service response timeout threshold and concurrent interactive session upper limit.

[0083] Optionally, update the interaction process control parameters according to the interaction response rules. For example, if the rules stipulate that the maximum response time for the payment page is 3 seconds, and the current service response timeout threshold is 5 seconds, update it to 3 seconds to ensure that the user waiting time is within a reasonable range. At the same time, adjust the user request routing strategy based on business needs and the current system load. For example, direct some user requests to server nodes with lower loads for processing; or adjust the upper limit of concurrent interaction sessions, for example, from 100 concurrent sessions to 150 to meet the needs of more users performing payment operations simultaneously. Through these updates, an updated interaction process control strategy is generated to optimize the interaction process between users and the system.

[0084] Step S148: Collect the optimized service indicators of the consumer service system in real time, calculate the matching degree between the optimized service indicators and the preset service optimization goals, and generate a service optimization deviation coefficient.

[0085] After optimizing the consumer service system, various service metrics are collected in real time. For example, the actual response time and success rate of the payment page are collected. These optimized service metrics are then compared with the pre-set service optimization targets. For example, if the preset average response time target for the payment page is 2 seconds, and the actual average response time collected is 3 seconds, a service optimization deviation coefficient is generated through corresponding calculation methods (such as calculating the ratio of the difference to the target value). This coefficient reflects the gap between the current optimization effect and the target, providing a basis for subsequent calibration.

[0086] Step S149: performing feedback calibration processing on the service adjustment parameter mapping table according to the service optimization deviation coefficient to generate a calibrated service adjustment parameter mapping table, and performing iterative optimization processing on the consumer service system using the calibrated service adjustment parameter mapping table.

[0087] Optionally, the service adjustment parameter mapping table is adjusted based on the service optimization deviation coefficient. If the deviation coefficient shows that the payment page response time is too long, it means that the current resource allocation and interactive response rules may need further optimization. Then, based on the deviation situation, the number of service nodes can be appropriately increased or the computing resource quota can be further adjusted. At the same time, parameters such as the service response timeout threshold may be adjusted again to generate a calibrated service adjustment parameter mapping table. This calibrated mapping table is then used to iteratively optimize the consumer service system again, continuously repeating the above-mentioned resource allocation, parameter update, indicator collection, and calibration process to gradually improve the interactive process performance of the consumer service system and continuously approach the preset service optimization target.

[0088] In an independently implementable technical solution, after optimizing the interaction process of the consumer service system corresponding to the target consumption scenario by using the consumer service dynamic adjustment strategy, the solution further includes any one of step S210, step S220, and step S230:

[0089] Step S210: Real-time collection of the interactive process optimization indicator set of the consumer service system, the interactive process optimization indicator set including the service response delay rate, resource allocation balance and user feedback emotional tendency change rate; dynamic weight allocation processing is performed on the interactive process optimization indicator set to generate an indicator weight distribution vector; wherein, the dynamic weight allocation processing adjusts the weight value based on the real-time fluctuation amplitude of the indicator and the historical optimization influence coefficient; based on the indicator weight distribution vector, the service resource configuration parameters in the dynamic adjustment strategy of the consumer service are gradient updated to generate updated service resource configuration parameters; wherein, the gradient update processing adjusts the parameter update step according to the degree to which the indicator deviates from the preset optimization threshold; the updated service resource configuration parameters are dynamically matched and calculated with the real-time load characteristics of the consumer service system to generate a secondary optimization strategy, and the interactive process control parameters are incrementally iteratively optimized based on the secondary optimization strategy.

[0090] After the initial interaction process optimization in the consumer service system, a set of interaction process optimization indicators are collected in real time. For example, system monitoring tools are used to obtain the service response delay rate, which is the ratio of the difference between the actual response time and the ideal response time. Resource allocation balance is calculated by analyzing resource allocation data to measure the uniformity of resource allocation across each service link. By collecting new user feedback and comparing it with previous feedback, the change rate of user feedback sentiment is calculated to understand the changes in user emotional attitudes towards the optimized service.

[0091] Next, these indicators are dynamically weighted. If the service response delay rate fluctuates significantly at a particular moment, and historical data shows that this metric has a significant impact on user experience, then a higher weight is assigned. Conversely, if the resource allocation balance is relatively stable and has a minimal impact on the overall optimization effect, a lower weight is assigned. This method generates an indicator weight distribution vector.

[0092] Based on this vector, the service resource configuration parameters are updated in a gradient manner. For example, if the service response delay rate has a high weight and deviates significantly from the preset optimization threshold, the service resource configuration parameters related to response time are adjusted in larger steps, such as adding more server resources to process requests.

[0093] The updated service resource configuration parameters are then dynamically matched with the real-time load characteristics of the consuming service system. For example, if the current real-time load indicates low user traffic and the updated resource configuration parameters are likely too conservative, the parameters are readjusted based on the load, generating a secondary optimization strategy.

[0094] Finally, we use a secondary optimization strategy to incrementally iterate and optimize the interaction process control parameters. For example, we gradually adjust parameters such as user request routing strategies and service response timeout thresholds, each with a small adjustment. This continuously improves the interaction process of the consumer service system, allowing it to continuously adapt to real-time conditions and approach optimal performance.

[0095] Step S220: Obtain the user real-time interaction behavior data after optimization of the consumer service system, the user real-time interaction behavior data including session interruption frequency, service option click distribution and interaction path jump sequence; perform multi-dimensional correlation analysis on the user real-time interaction behavior data to generate a behavior pattern feature matrix; wherein the multi-dimensional correlation analysis includes jointly encoding the temporal correlation of the behavior sequence in the same user session and the cross-session behavior consistency; perform cross-attention calculation on the behavior pattern feature matrix and the service optimization feature set to generate behavior-service correlation features; the cross-attention calculation filters the strong correlation dimensions of the behavior pattern and the service optimization features through the attention weight matrix; according to the behavior-service correlation features, perform adaptive correction processing on the user request routing strategy in the consumer service dynamic adjustment strategy to generate a real-time routing priority queue based on behavior prediction, and update the interaction response timeout threshold based on the queue.

[0096] After optimizing the consumer service system, we collect real-time user interaction behavior data. For example, by embedding monitoring code in the system, we can record the frequency of each user's session interruption, that is, the number of times a user unexpectedly interrupts their shopping session. We can also count the distribution of service option clicks to understand user preferences for different service options (such as payment methods and delivery options). We can also record the jump sequence of the interaction path to understand the order and frequency of users' jumps between different pages.

[0097] Perform multi-dimensional correlation analysis on this data. Within the same user session, analyze the temporal correlation of behavioral sequences, such as how long after clicking the payment option the user confirms the action, and what the temporal relationship is between them. Also consider cross-session behavioral consistency, namely, whether users exhibit similar behavioral patterns across different shopping sessions. This information is integrated through joint encoding to generate a behavioral pattern feature matrix.

[0098] Cross-attention calculations are performed on the behavioral pattern feature matrix and the service optimization feature set. The attention weight matrix identifies strongly correlated dimensions based on the correlation between behavioral patterns and service optimization features. For example, if users frequently click on a specific payment method option, and the payment process for that option has room for improvement (corresponding to a service optimization feature), this dimension can be considered a strongly correlated dimension.

[0099] Based on the generated behavior-service correlation features, the user request routing strategy is adaptively modified. For example, if a certain user's behavior pattern indicates a preference for quick payment, requests from these users are prioritized to server nodes with faster processing speeds, generating a real-time routing priority queue based on behavior prediction. Simultaneously, the interaction response timeout threshold is updated based on this queue. For users who prefer fast payment, the response timeout threshold is appropriately shortened to improve their interaction experience, ensuring that the system can dynamically adjust interaction process control parameters based on user behavior.

[0100] Step S230: Synchronously obtain cross-system service status data associated with the target consumption scenario, the cross-system service status data including the success rate of external service interface calls, third-party resource scheduling delays, and cross-platform user identification mapping relationships; perform service link topology analysis on the cross-system service status data to generate a cross-system dependency graph; wherein, the service link topology analysis constructs dependency weights through the call frequency between nodes and the data flow transmission delay; based on the cross-system dependency graph, perform global optimization impact prediction on the service optimization feature set to generate cross-system optimization constraints; the prediction process calculates the propagation influence coefficient of the service optimization feature in the dependency graph through a graph neural network; according to the cross-system optimization constraints, perform collaborative adjustment on the service thread priority in the dynamic adjustment strategy of the consumer service to generate a cross-system resource allocation strategy, and dynamically reconstruct the computing resource quota and concurrent session limit of the service node based on the strategy.

[0101] After optimizing the interaction process within the consumer service system, synchronize cross-system service status data. For example, consider the success rate of external service interface calls to third-party payment platforms that collaborate with e-commerce platforms. A low success rate can impact the user's payment process. Delays in third-party resource scheduling, such as logistics resource scheduling, can impact product delivery services. Furthermore, cross-platform user identity mapping relationships are used to understand the user's identity associations across different platforms, which can impact the user experience when switching between platforms.

[0102] This data is used to perform service link topology analysis. Dependency weights are constructed by counting the call frequency between nodes, such as the number of calls between the e-commerce platform and the third-party payment platform, as well as data flow transmission latency. If the e-commerce platform frequently calls the third-party payment platform and the transmission latency is high, the dependency weight between these two nodes in the dependency graph will be large. This generates a cross-system dependency graph that clearly shows the connections and dependencies between various systems.

[0103] Graph neural networks are used to calculate the propagation impact of service optimization feature sets across a cross-system dependency graph. For example, this can analyze how optimizing the payment page interaction process might impact related third-party payment platforms and other dependent systems. Based on the calculation results, cross-system optimization constraints are generated to clarify the limiting factors that need to be considered when optimizing across systems.

[0104] Based on these constraints, the service thread priorities in the consumer service dynamic adjustment strategy are collaboratively adjusted. For example, if it is found that the processing capacity of the third-party payment platform has a significant impact on the overall shopping process, and the current service thread priority setting is not conducive to quickly calling the platform interface, then the priority of the relevant service thread will be increased.

[0105] Based on the results of the collaborative adjustments, a cross-system resource allocation strategy is generated. For example, to ensure a smooth payment process, the computing resource quota of service nodes related to interaction with third-party payment platforms is appropriately increased. At the same time, based on cross-system dependencies and user access volume, the concurrent session limit is dynamically restructured to ensure that services across the entire cross-system environment can operate efficiently and provide users with a stable and smooth consumer interaction experience.

[0106] In practical applications, the semantic encoding model aims to conduct in-depth analysis of the text unit sequence formed after text unit division in the entire consumer feedback processing process, so as to generate contextual semantic features that can accurately reflect the semantic picture of the user feedback sentence.

[0107] The input to the semantic encoding model is an encoding input vector obtained after a series of processing. The generation process of this vector is relatively complex. First, the system maps each text unit in the text unit sequence into a vector space to generate an initial text unit vector. These vectors represent the position and characteristics of the text unit in the semantic space. Next, considering the position of the text unit in the sequence, the system constructs a position encoding feature based on the positional correlation of adjacent text units and superimposes it with the initial text unit vector to obtain the encoding input vector. This vector contains both the semantic information of the text unit itself and its position information, providing a comprehensive data foundation for subsequent calculations of the semantic encoding model.

[0108] Within the semantic encoding model, semantic attention calculations are performed on the encoded input vector at multiple levels. At each semantic attention layer, the model first performs multi-head attention calculations, focusing on the encoded input vector from different perspectives. For example, different heads focus on the keyword portion and the overall structural information in the vector, respectively, to generate intermediate attention features. Subsequently, the intermediate attention features are residually connected to the encoded input vector. This connection method preserves important information in the original vector, resulting in residual features. Next, the residual features are layer-normalized to stabilize the data distribution, resulting in normalized features. The normalized features are then input into a feedforward neural network for nonlinear transformation, increasing the model's expressive power, resulting in the output features of the current semantic attention layer. The model iteratively performs multiple layers of semantic attention calculations until the preset number of layers is reached, ultimately outputting the output features of the last layer as contextual semantic features that contain global semantic relevance.

[0109] Through the above-mentioned input and output and connection relationship settings at each layer, the semantic encoding model can deeply analyze the semantic relevance and weight distribution between text units. The generated contextual semantic features can accurately reflect the overall semantic information of user feedback sentences, providing a solid semantic foundation for subsequent steps such as sentiment analysis and service optimization feature generation, helping e-commerce platforms better understand user feedback and thus optimize consumer service systems.

[0110] It is understandable that in actual application, when implementing the above technical solutions, technical personnel in the relevant field can define the minimum data unit in the time series sentiment analysis processing as a time window containing at least 3 consecutive user feedbacks in a single session based on the sliding window mechanism and time series segmentation method in the existing technology, and reasonably divide the interaction records by setting fixed time intervals or event trigger thresholds, so as to ensure the statistical significance and business interpretability of the sentiment fluctuation trend analysis.

[0111] In addition, for the calculation of historical optimization influence coefficients in dynamic weight allocation processing, a historical strategy effect attenuation model can be constructed based on the temporal difference error update mechanism in reinforcement learning. The strategy execution effect in the past 30 days can be dynamically attenuated by the exponentially weighted moving average method, and a normalized influence coefficient can be generated in combination with the current service load status, thereby achieving adaptive adjustment of the indicator weights.

[0112] Furthermore, to address the dimensionality unification problem of service resource configuration parameters, the Min-Max normalization and Z-score standardization combination method widely used in the industry can be adopted to perform phased standardization on heterogeneous parameters such as the number of service nodes and computing resource quotas: first, each parameter is linearly mapped to the [0, 1] interval through Min-Max to eliminate dimensional differences, and then Gaussian normalization is performed using Z-score (μ=0, σ=1). Finally, a service resource configuration vector with uniform distribution characteristics is generated, thereby ensuring the convergence stability of the dynamic resource allocation algorithm in the multidimensional parameter space.

[0113] In cross-system dependency analysis, knowledge graph embedding technology and graph attention network (GAT) can be combined to jointly encode node attributes and edge weights in the service link topology into a low-dimensional vector representation, and deep correlation features of cross-system service status can be extracted through graph convolution operations, thereby improving the accuracy of global optimization impact prediction.

[0114] For user behavior pattern analysis, the Transformer-XL model can be introduced to perform segmented recursive modeling of long sequence interaction paths, use the relative position encoding mechanism to capture cross-session behavior correlation, and combine the contrastive learning framework to construct a user behavior representation space, thereby enhancing the interpretability of behavior-service association features.

[0115] In this way, the robustness and scalability of the dynamic adjustment strategy of consumer services can be significantly improved, thereby adapting to the optimization needs of complex e-commerce scenarios with high concurrency, multi-dimensionality, and cross-platform.

[0116] It should be noted that the acquisition and processing of user feedback statements and related user behavior data involved in this embodiment strictly comply with relevant laws, regulations and industry standards for personal information protection. In specific implementation, the collection of all user feedback information is carried out under the premise that the user is fully informed and explicitly authorized. When submitting feedback, the user can choose whether to agree to their data being used for service optimization analysis, and support the withdrawal of authorization at any time. For the collected raw data, the system uses de-identification technology to anonymize the user's identity information to ensure that personal identity cannot be reversely identified through feedback content or behavioral characteristics. The processed data is only used to enhance the interactive experience and function optimization of the consumer service system, and is prohibited from being used in any scenario unrelated to the authorized purpose. The data storage and transmission process are all protected by encryption algorithms and access control mechanisms to effectively protect the privacy rights and interests of users. This embodiment always adheres to the data use principles of "minimum necessary, informed consent, safe and controllable" to ensure the organic unity of technology application and privacy protection.

[0117] In summary, this embodiment comprehensively collects various types of information from the consumption process by acquiring a consumer feedback text collection containing multiple consumer interaction records in the target consumption scenario. Dynamic semantic analysis of the consumer feedback text collection accurately extracts the semantic and sentiment feature sets of each consumer interaction record to deeply explore the connotation of user feedback. Based on the dynamic correlation mapping processing of semantic and sentiment feature sets, a targeted service optimization feature set can be generated, providing strong support for accurately identifying service improvement directions. Dynamic adjustment strategies for consumer services are then generated, and interaction processes of the consumer service system are optimized, effectively improving the adaptability and flexibility of the consumer service system.

[0118] See also Figure 2 As shown in FIG. 1 , this figure is a schematic diagram of the basic structure of a computer system 200 provided in an embodiment of the present invention. The computer system 200 includes:

[0119] Processor 201;

[0120] a storage device 202 having a computer program 2020 stored thereon;

[0121] When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the dynamic analysis methods for consumption big data.

[0122] Based on the above, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0123] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems or devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

Claims

1. A dynamic analysis method for consumption big data, characterized in that: include: Obtaining a consumer feedback text set in a target consumption scenario, wherein the consumer feedback text set includes multiple consumer interaction records, each consumer interaction record consisting of at least one user feedback statement and a corresponding consumer service identifier; Performing dynamic semantic analysis on the consumer feedback text set to obtain a semantic feature set and a sentiment tendency feature set of each consumer interaction record; Generating a service optimization feature set for the consumer interaction record based on a dynamic association mapping process between the semantic feature set and the emotional tendency feature set; Generating a consumption service dynamic adjustment strategy based on the service optimization feature set, and using the consumption service dynamic adjustment strategy to optimize the interaction process of the consumption service system corresponding to the target consumption scenario; Generating a consumption service dynamic adjustment strategy according to the service optimization feature set includes: Performing cluster analysis on the service optimization feature set to generate multiple service optimization categories and corresponding category center features; Perform matching retrieval based on the category center feature and the historical optimization strategy library to determine the initial service adjustment strategy set; Prioritizing the initial service adjustment policy set according to the real-time consumption service load status to generate a policy execution sequence; Taking the subset of strategies that meet the real-time load constraint in the strategy execution sequence as the consumption service dynamic adjustment strategy; The utilizing the dynamic adjustment strategy of the consumption service to optimize the interaction process of the consumption service system corresponding to the target consumption scenario includes: Obtaining a service adjustment parameter mapping table in the consumption service dynamic adjustment strategy, wherein the service adjustment parameter mapping table includes service resource configuration parameters and interactive response rules corresponding to multiple service optimization categories; Performing dynamic resource allocation processing on the real-time service load indicator of the consumer service system according to the service resource configuration parameters to generate a resource allocation optimization result; wherein the dynamic resource allocation processing includes adjusting at least one of the number of service nodes, computing resource quota, and service thread priority; Based on the interactive response rules, the interactive process control parameters of the consumer service system are updated to generate an updated interactive process control policy; the interactive process control parameters include at least one of the following: user request routing policy, service response timeout threshold, and concurrent interactive session upper limit; Collecting optimized service indicators of the consumer service system in real time, calculating the matching degree between the optimized service indicators and preset service optimization targets, and generating a service optimization deviation coefficient; Feedback calibration processing is performed on the service adjustment parameter mapping table according to the service optimization deviation coefficient to generate a calibrated service adjustment parameter mapping table, and the calibrated service adjustment parameter mapping table is used to iteratively optimize the consumer service system.

2. The dynamic analysis method of consumption big data according to claim 1 is characterized in that: The dynamic semantic analysis processing of the consumer feedback text set to obtain a semantic feature set and a sentiment tendency feature set of each consumer interaction record includes: Performing text unit division processing on the user feedback sentences in the consumption interaction record to obtain a plurality of semantically associated text unit sequences; Calling a pre-trained semantic encoding model to perform context encoding processing on the text unit sequence to generate contextual semantic features of the user feedback sentence; wherein the contextual semantic features include semantic associations and weight distributions between different text units in the text unit sequence; Performing sentiment polarity recognition processing on the user feedback sentence to generate sentiment tendency features of the user feedback sentence; the sentiment tendency features include at least one of the following: sentiment polarity intensity, sentiment category distribution, and sentiment fluctuation trend of the user feedback sentence; The context semantic features and the sentiment tendency features are subjected to feature fusion processing to obtain a semantic feature set and a sentiment tendency feature set of the consumption interaction record.

3. The dynamic analysis method of consumption big data according to claim 2 is characterized in that: The calling of a pre-trained semantic encoding model to perform context encoding processing on the text unit sequence to generate contextual semantic features of the user feedback sentence includes: Performing vectorization conversion processing on each text unit in the text unit sequence to generate an initial text unit vector; constructing a position encoding feature of the initial text unit vector based on the position correlation between adjacent text units in the text unit sequence; Superimposing the initial text unit vector and the position coding feature to obtain a coding input vector; The semantic encoding model is called to perform multi-layer semantic attention calculation on the encoded input vector to generate contextual semantic features containing global semantic relevance.

4. The dynamic analysis method of consumption big data according to claim 3 is characterized in that: The calling of the semantic encoding model to perform multi-layer semantic attention calculation on the encoded input vector to generate contextual semantic features containing global semantic relevance includes: In each semantic attention layer of the semantic encoding model, performing multi-head attention calculation on the encoding input vector to generate intermediate attention features; Performing a residual connection process on the intermediate attention feature and the encoded input vector to obtain a residual feature; Performing layer normalization processing on the residual features to obtain normalized features; Inputting the normalized features into a feedforward neural network for nonlinear transformation to obtain the output features of the current semantic attention layer; Iteratively perform multi-layer semantic attention calculation until a preset number of layers is reached, and the output features of the last layer are used as the contextual semantic features.

5. The dynamic analysis method of consumption big data according to claim 2 is characterized in that: The performing sentiment polarity recognition processing on the user feedback sentence to generate the sentiment tendency feature of the user feedback sentence includes: Extracting emotional keywords from the user feedback sentences to obtain an emotional keyword set and corresponding emotional weights; Based on the sentiment knowledge base and the sentiment weight, performing sentiment polarity labeling processing on the sentiment keyword set to generate a sentiment polarity distribution vector; Performing temporal sentiment analysis on the user feedback sentence to generate sentiment fluctuation sequence features; wherein the sentiment fluctuation sequence features reflect the temporal variation pattern of sentiment polarity in the user feedback sentence; The emotion polarity distribution vector and the emotion fluctuation sequence feature are jointly encoded to generate the emotion tendency feature.

6. The method for dynamic analysis of consumption big data according to claim 5, characterized in that: The performing time-series sentiment analysis on the user feedback sentence to generate sentiment fluctuation sequence features includes: Dividing the user feedback sentence into multiple sentiment analysis windows in chronological order; In each sentiment analysis window, the sentiment classification model is called to predict the sentiment polarity of the text content in the current sentiment analysis window and generate a window sentiment label; Based on the sliding window mechanism, the emotional labels of adjacent windows are differentially calculated to generate the emotional label change rate sequence; A trend fitting process is performed on the emotion label change rate sequence to generate emotion fluctuation sequence features reflecting the direction and intensity of emotion fluctuation.

7. The method for dynamic analysis of consumption big data according to claim 1, characterized in that: The generating of the service optimization feature set of the consumption interaction record based on the dynamic association mapping process between the semantic feature set and the emotional tendency feature set includes: Calculating importance weights of contextual semantic features in the semantic feature set to generate an importance weight distribution; Performing dynamic weighted fusion processing on the emotional tendency feature set according to the importance weight distribution to generate a joint emotional semantic feature; Performing an association mapping process on the joint emotional semantic feature and the consumer service identifier to generate a service association feature; wherein the service association feature reflects the matching degree and optimization priority between the user feedback statement and the consumer service identifier; Performing dimensionality reduction and compression processing on the service-related features to generate the service optimization feature set.

8. The method for dynamic analysis of consumption big data according to claim 7, characterized in that: The step of performing an associative mapping process on the combined emotional semantic feature and the consumer service identifier to generate a service-associated feature includes: Obtaining a service attribute feature set corresponding to the consumer service identifier; the service attribute feature set includes at least one of the following: a service response timeliness feature, a service coverage feature, and a service resource allocation feature; Calculating a multidimensional similarity matrix between the joint sentiment semantic feature and the service attribute feature set; Performing feature screening on the multidimensional similarity matrix to retain associated dimensions that meet a preset similarity threshold; A service association feature between the user feedback statement and the consumption service identifier is constructed based on the retained association dimension.

9. A computer system, characterized in that: include: processor; A storage device having a computer program stored thereon, wherein when the computer program is executed by the processor, the processor implements the dynamic analysis method of consumption big data as described in any one of claims 1-8.

Images