Big data recommendation method and big data mining system applied to cloud digital business

Abstract

The present invention relates to the fields of digital economy, artificial intelligence and big data technology, and provides a big data recommendation method and a big data mining system applied to cloud digital services, which can obtain coding vectors of topic interest focus points that are as comprehensive and complete as possible and can characterize relevant feature semantics. In particular, when the selected online topic interaction scenario affects the user session activity record to be analyzed, resulting in the missing of the mined viewpoint comment quantization frequent items and low feature recognition, the viewpoint comment quantization frequent items are optimized according to the target comment keyword features to help capture the topic interest, and ensure the integrity and feature recognition of the obtained target session text quantization coding. The topic interest capture information obtained by capturing the topic interest of the user session activity record to be analyzed based on the target session text quantization coding is more accurate and reasonable, which improves the accuracy and credibility of the topic interest capture, and provides a more credible analysis basis for the big data recommendation of the digital service industry.

Description

Technical Field

[0001] The present invention relates to the fields of digital economy, artificial intelligence and big data technology, and in particular to a big data recommendation method and a big data mining system applied to cloud digital business. Background Art

[0002] With the rapid development of digital technologies such as the Internet, big data, cloud computing, artificial intelligence, and blockchain, the digital economy has increasingly become a key force in the reorganization of global industrial resources, structural reshaping, and the evolution of the competitive landscape. The digital economy is a key force driving the high-quality development of the industry, and high-quality industrial development is an important support for the development of the digital economy. High-quality industrial development refers to the advanced state of industrial development quality. Its process is inevitably accompanied by the upgrading and adjustment of the industrial structure and the accelerated emergence of emerging formats. Its core measurement dimensions include industrial competitiveness, industrial innovation, industrial sustainable development capabilities, and industrial security.

[0003] The core logic of the digital economy driving the high-quality development of the industry includes multiple mechanisms such as the production factor empowerment mechanism, transaction cost reduction mechanism, industrial organization transformation mechanism, and resource allocation capability improvement. Specifically, digital factors become the core production factors for the high-quality development of the industry, digital open innovation systems and user innovation become the new normal for cross-industry integration and innovation, cross-industry chain integration under the organization of industrial platforms, and cross-niche collaborative industrial ecological chain groups become new industrial organizations for the co-creation of industrial chain value, and the high flexibility of the digital industrial chain improves the resilience and risk resistance of the industrial chain. Summary of the invention

[0004] The present invention at least provides a big data recommendation method and a big data mining system applied to cloud digital services.

[0005] The technical solution of the present invention is achieved through at least some of the following embodiments.

[0006] A big data recommendation method applied to cloud digital business, applied to a big data mining system, the method comprising:

[0007] Obtaining the crawled user session activity records to be analyzed in the selected online topic interaction scenario; performing conversation opinion and comment mining on the user session activity records to be analyzed to obtain quantified frequent items of opinion and comment;

[0008] Based on the commonality measurement results of the frequent quantified items of the opinion comments and the comment keyword features in the comment keyword feature pool, a target comment keyword feature that is related to the frequent quantified items of the opinion comments is determined from the comment keyword feature pool, wherein the comment keyword feature pool includes comment keyword features with different feature semantics, and the comment keyword feature is a text detail representation vector corresponding to the topic interest focus points with different feature semantics;

[0009] The quantitative frequent items of the opinion comments are optimized according to the keyword features of the target comments to obtain the current opinion comment frequent items; the target conversation text quantitative coding corresponding to the user conversation activity record to be analyzed is determined based on the current opinion comment frequent items; the topic interest of the user conversation activity record to be analyzed is captured based on the target conversation text quantitative coding to obtain topic interest capture information, so as to guide the big data recommendation of the digital service business of the digital service industry corresponding to the selected online topic interaction scenario based on the topic interest capture information.

[0010] In some exemplary embodiments, the method further comprises:

[0011] Loading the user session activity record to be analyzed into a topic interest capture network, wherein the topic interest capture network includes a conversation viewpoint comment mining component, a comment keyword feature component, a vector aggregation component and a topic interest capture component, wherein the comment keyword feature component is formed by comment keyword features in the comment keyword feature pool;

[0012] The step of mining the conversation viewpoints and comments of the user session activity record to be analyzed to obtain the viewpoint and comments quantified frequent items includes: mining the conversation viewpoints and comments of the user session activity record to be analyzed according to the conversation viewpoint and comments mining component to obtain the viewpoint and comments quantified frequent items;

[0013] The method of determining a target comment keyword feature associated with the opinion comment quantification frequent item from the comment keyword feature pool based on the commonality measurement result of the opinion comment quantification frequent item and the comment keyword feature in the comment keyword feature pool comprises: determining the target comment keyword feature associated with the opinion comment quantification frequent item based on the comment keyword feature component;

[0014] The step of optimizing the viewpoint review quantified frequent item according to the target review keyword feature to obtain the current viewpoint review frequent item includes: aggregating the target review keyword feature and the viewpoint review quantified frequent item according to the vector aggregation component to obtain the current viewpoint review frequent item;

[0015] The determining the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion and comment frequent items includes: determining the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion and comment frequent items according to the vector aggregation component;

[0016] The topic interest capturing of the user conversation activity record to be analyzed based on the quantization encoding of the target conversation text to obtain the topic interest capturing information includes: the topic interest capturing of the user conversation activity record to be analyzed based on the quantization encoding of the target conversation text according to the topic interest capturing component to obtain the topic interest capturing information.

[0017] In some exemplary embodiments, the method further includes generating the comment keyword feature component, wherein generating the comment keyword feature component includes:

[0018] In the process of debugging the topic interest capture network, the user session activity record sample is mined for dialogue opinion and comment according to the dialogue opinion and comment mining component to obtain the quantized frequent item samples of the opinion and comment of the topic interest focus matched by the prior feature semantic annotation, wherein the prior feature semantic annotation is used to mark the feature semantics of the topic interest focus included in the user session activity record sample;

[0019] The comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component are adjusted based on the quantified frequent item samples of the opinion comments on the topic interest focus.

[0020] In some exemplary embodiments, the step of adjusting the comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component based on the frequent item examples of the opinion comments quantified on the topic interest focus includes:

[0021] If the number of comment keyword features of the characteristic semantics corresponding to the topic interest focus point is 0, then the quantified frequent item sample of the opinion comment of the topic interest focus point is used as the comment keyword feature of the characteristic semantics corresponding to the topic interest focus point;

[0022] If the number of the comment keyword features of the characteristic semantics corresponding to the topic interest focus point is ≠ 0, then determining the commonality measurement result between the quantized frequent item samples of the opinion comments of the topic interest focus point and the comment keyword features of the characteristic semantics corresponding to the topic interest focus point;

[0023] The comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component are adjusted based on the commonality measurement result.

[0024] In some exemplary embodiments, adjusting the comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component based on the commonality measurement result includes:

[0025] If the commonality measurement result is less than the commonality measurement result limit, the quantized frequent item samples of the opinions and comments on the topic interest focus are added to the comment keyword features of the feature semantics corresponding to the topic interest focus.

[0026] In some exemplary embodiments, adjusting the comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component based on the commonality measurement result includes:

[0027] If the commonality measurement result is less than the commonality measurement result limit, determining a comparison result between the number of comment keyword features of the feature semantics corresponding to the topic interest focus and the number limit;

[0028] The comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component are adjusted based on the comparison result.

[0029] In some exemplary embodiments, generating the comment keyword feature component includes:

[0030] Generate a target debugging cost indicator based on the difference between the prior feature semantic annotation and the obtained current opinion comment frequent item sample;

[0031] The comment keyword features in the comment keyword feature component are adjusted based on the target debugging cost indicator to generate the comment keyword feature component.

[0032] In some exemplary embodiments, different online topic interaction scenarios correspond to different comment keyword feature pools, and determining a target comment keyword feature associated with the opinion comment quantification frequent item from the comment keyword feature pool based on the commonality measurement result between the opinion comment quantification frequent item and the comment keyword feature in the comment keyword feature pool includes:

[0033] Obtaining a comment keyword feature pool corresponding to the selected online topic interaction scenario;

[0034] The target comment keyword feature that is associated with the quantified frequent item of the opinion comment is determined from the comment keyword feature pool corresponding to the selected online topic interaction scenario.

[0035] In some exemplary embodiments, determining the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion and comment frequent items includes: combining the current opinion and comment frequent items with the opinion and comment quantization frequent items to obtain the target conversation text quantization code.

[0036] In some exemplary embodiments, optimizing the opinion review quantified frequent items according to the target review keyword feature to obtain the current opinion review frequent items includes:

[0037] Calculating a commonality measurement result trust list based on the target comment keyword feature and the opinion comment quantization frequent item, wherein the members in the commonality measurement result trust list are used to mark the importance coefficient of each target comment keyword feature to the opinion comment quantization frequent item;

[0038] Based on the commonality measurement result credible list, the target comment keyword features and the opinion comment quantification frequent items are enhanced and integrated to obtain the current opinion comment frequent items corresponding to the opinion comment quantification frequent items.

[0039] In some exemplary embodiments, the method further comprises:

[0040] Obtaining a user session activity record sample under an online topic interaction scenario sample, wherein the user session activity record sample has a priori feature semantic annotations of the topic interest points included therein;

[0041] The user session activity record sample is loaded into a general topic interest capture network, wherein the general topic interest capture network includes a general dialogue opinion comment mining component, a general comment keyword feature component, a general vector aggregation component and a general topic interest capture component, and the user session activity record sample has a prior feature semantic annotation of the topic interest focus included therein;

[0042] Performing conversation opinion and comment mining on the user session activity record sample according to the general conversation opinion and comment mining component to obtain opinion and comment quantized frequent item samples;

[0043] According to the general comment keyword feature component, based on the commonality measurement results of the opinion comment quantization frequent item sample and the comment keyword features in the comment keyword feature pool, determine the target comment keyword feature sample that is related to the opinion comment quantization frequent item sample from the general comment keyword feature component;

[0044] According to the general vector aggregation component, the target comment keyword feature sample is used to optimize the opinion comment quantized frequent item sample to obtain the current opinion comment frequent item sample;

[0045] According to the general vector aggregation component, based on the current opinion review frequent item sample, determine the target conversation text quantization encoding sample corresponding to the user conversation activity record sample;

[0046] According to the general topic interest capture component, based on the target conversation text quantization coding sample, the topic interest of the user conversation activity record to be analyzed is captured to obtain a topic interest capture information sample;

[0047] The general topic interest capture network is debugged based on the prior feature semantic annotation and the topic interest capture information sample to obtain the topic interest capture network.

[0048] A big data mining system comprises: a processor, a memory and a network interface; the processor is connected to the memory and the network interface; the network interface is used to provide data communication function, the memory is used to store program code, and the processor is used to call the program code to execute the above method.

[0049] A computer-readable storage medium stores a computer program, which executes a big data recommendation method applied to cloud digital business when running.

[0050] A computer program product includes a computer program or computer executable instructions, which, when executed by a processor, implements a big data recommendation method applied to cloud digital services.

[0051] According to an embodiment of the present invention, the present invention counts the comment keyword features of different feature semantics in advance, and the comment keyword features are the text detail representation vectors corresponding to the topic interest focus points of different feature semantics, which can represent the topic interest of a feature semantics. When capturing the topic interest of the crawled user session activity record to be analyzed in the selected online topic interaction scenario, the dialogue viewpoint comment mining can be performed on the user session activity record to be analyzed to obtain the viewpoint comment quantified frequent items, and then based on the commonality measurement results of the viewpoint comment quantified frequent items and the comment keyword features in the comment keyword feature pool, the target comment keyword features that are related to the viewpoint comment quantified frequent items are determined from the comment keyword feature pool. The comment keyword feature pool includes comment keyword features of different feature semantics, and the comment keyword features are the text detail representation vectors corresponding to the topic interest focus points of different feature semantics. Therefore, the viewpoint comment quantified frequent items can be optimized based on the target comment keyword features to obtain the current viewpoint comment frequent items, and the target conversation text quantified encoding corresponding to the user session activity record to be analyzed can be determined based on the current viewpoint comment frequent items, so as to obtain the encoding vector that is as comprehensive and complete as possible and can represent the topic interest focus points of the relevant feature semantics. In particular, when the selected online topic interaction scenario affects the user conversation activity records to be analyzed, resulting in the missing of the mined opinion and comment quantification frequent items and low feature recognition, the opinion and comment quantification frequent items are optimized according to the target comment keyword features to help capture the topic interest, and ensure the integrity and feature recognition of the obtained target conversation text quantification coding. In this way, the topic interest of the user conversation activity records to be analyzed is captured based on the target conversation text quantification coding, and the obtained topic interest capture information is more accurate and reasonable, which improves the accuracy and credibility of the topic interest capture, and provides a more credible analysis basis for the big data recommendation of digital service businesses in the relevant digital service industry. In summary, the embodiments of the present invention improve the technical problem that the prior art is difficult to accurately and reliably realize topic interest capture, and thus improve the technical problem that it is difficult to provide a more credible analysis basis for the big data recommendation of digital service businesses in the relevant digital service industry.

[0052] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present invention and are used together with the specification to illustrate the technical solutions of the present invention. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 A schematic diagram of a big data mining system provided by an embodiment of the present invention is shown.

[0055] Figure 2 A flow chart of a big data recommendation method applied to cloud digital services provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention described and shown herein can be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0058] Figure 1 A schematic diagram of a big data mining system provided by an embodiment of the present invention is shown, and the big data mining system 100 includes a processor 110, a memory 120, and a network interface 130. The processor 110 is connected to the memory 120 and the network interface 130. Further, the network interface 130 is used to provide a data communication function, the memory 120 is used to store program code, and the processor 110 is used to call the program code to execute a big data recommendation method applied to cloud digital services.

[0059] Figure 2 is a flow chart showing a big data recommendation method applied to cloud digital services that can implement an embodiment of the present invention. The big data recommendation method applied to cloud digital services can be implemented by Figure 1The big data mining system 100 shown implements a big data recommendation method applied to cloud digital services, illustratively including steps 201 to 206.

[0060] Step 201: Obtain the crawled user session activity records to be analyzed in the selected online topic interaction scenario.

[0061] In actual implementation, the text big data crawler may crawl user session activity records (texts) in different online topic interaction scenarios. However, the network state stability of these online topic interaction scenarios may be poor, and the crawled user session activity records (texts) may be affected by such online topic interaction scenarios, such as errors or omissions in the user session activity records (texts). According to the design ideas of the embodiments of the present invention, the user session activity records (texts) crawled in these online topic interaction scenarios can be used to capture topic interests, and obtain topic interest capture information as accurate as possible. In the embodiment of the present invention, the online topic interaction scenario is taken as an example of the selected online topic interaction scenario. The user session activity records (texts) crawled in the selected online topic interaction scenario for topic interest capture can be understood as user session activity records to be analyzed. Among them, the selected online topic interaction scenario can be an online topic interaction scenario for digital virtual space services (such as metaverse, VR games, VR malls, etc.). As a high-heat and high-traffic branch under the new digital economy, the online topic interaction scenario corresponding to the digital virtual space service has considerable mining value and guiding significance.

[0062] In the embodiment of the present invention, the user session activity records to be analyzed include a series of session interaction information of users of different online platforms, such as voice conversation information, text chat information, emoticon image information, etc.

[0063] Step 202: Mining the conversation opinions and comments of the user session activity record to be analyzed to obtain quantified frequent items of opinions and comments.

[0064] In an embodiment of the present invention, after obtaining the user session activity record to be analyzed, the user session activity record to be analyzed can be loaded into a topic interest capture network to capture the topic interest of the user session activity record to be analyzed through the topic interest capture network to obtain topic interest capture information.

[0065] Furthermore, the topic interest capture network includes a conversation opinion comment mining component, a comment keyword feature component, a vector aggregation component and a topic interest capture component. The comment keyword feature component is formed by the comment keyword features in the comment keyword feature pool. Using the topic interest capture network to capture the topic interest of the user session activity record to be analyzed can include two links. The first link is to mine the conversation opinion comment based on the topic interest capture network, such as mining the conversation opinion comment based on the conversation opinion comment mining component to obtain the frequent items of opinion comment quantification. The second link is mainly to perform subsequent processing based on the topic interest capture network through the frequent items of opinion comment quantification mined. The second link requires the application of the comment keyword feature component, the vector aggregation component and the topic interest capture component.

[0066] Furthermore, the topic interest capture network can be a neural network model trained and debugged based on artificial intelligence technology, such as a convolutional neural network, a deep structured semantic model, a residual network, etc. Dialogue opinion and comment mining is used to extract dialogue opinion and comment features, and the obtained opinion and comment quantization frequent items can be understood as quantized opinion and comment features. Among them, using opinion and comment quantization frequent items to characterize opinion and comment features can avoid excessive system computing overhead caused by floating-point features, and because topic analysis focuses on a certain subjective analysis, the granularity of some floating-point features can be sacrificed, thereby ensuring the timeliness of the overall solution.

[0067] Based on the above content, the conversation opinion and comment mining component used in the first stage may include an interest identification layer and a reversible layer. First, based on the interest identification layer of the first stage, the distribution window that may correspond to the topic interest focus in the user conversation activity record to be analyzed is obtained to obtain the interest identification window.

[0068] Then, these interest recognition windows are cleaned according to NMS (Non-Maximum Suppression) to remove the overlapping interest recognition windows for the same topic. These interest recognition windows obtain the corresponding frequent items of opinion comments according to the reversible layer (convolutional layer), and then enter the second stage for processing to obtain the final feature semantics and distribution window.

[0069] Step 203: Based on the commonality measurement results of the opinion and comment quantification frequent items and the comment keyword features in the comment keyword feature pool, determine the target comment keyword features associated with the opinion and comment quantification frequent items from the comment keyword feature pool.

[0070] In the embodiment of the present invention, a comment keyword feature pool can be generated in advance, and the comment keyword feature pool includes comment keyword features of different feature semantics (semantic types), and the comment keyword features are text detail representation vectors (significant feature vectors) corresponding to topic interest points of different feature semantics. The comment keyword feature pool in the embodiment of the present invention can form a comment keyword feature component, and the comment keyword feature component can also be understood as a class center component.

[0071] Among them, the comment keyword features in the comment keyword feature pool can be collected during the debugging process of the topic interest capture network. During the debugging process of the topic interest capture network, the text detail representation vector of the topic interest focus of each feature semantics is continuously collected as the comment keyword feature of this feature semantics. The comment keyword feature is not only used in the debugging process, but also used to provide a reference for topic interest capture.

[0072] Considering that the user session activity records to be analyzed may be affected by the selected online topic interaction scenario, the obtained frequent items of opinion and comment quantification may have low feature recognition, so based on the commonality measurement results of frequent items of opinion and comment quantification and comment keyword features in the comment keyword feature pool, target comment keyword features that are related to frequent items of opinion and comment quantification can be selected from the comment keyword feature pool. For example, the commonality measurement results (feature similarity) of frequent items of opinion and comment quantification and each comment keyword feature in the comment keyword feature pool are determined. The higher the commonality measurement result, the more likely it is that the topic interest focus corresponding to the comment keyword feature and the topic interest focus corresponding to the frequent items of opinion and comment quantification may be the same topic interest focus with the same feature semantics, so the comment keyword feature with a commonality measurement result higher than a certain limit can be used as the target comment keyword feature. Given that the comment keyword feature pool includes comment keyword features of different feature semantics, the comment keyword features are text detail representation vectors corresponding to topic interest and focus points of different feature semantics, and the determined target comment keyword features may be text detail representation vectors of topic interest and focus points corresponding to the quantitative frequent items of opinion comments. Therefore, the target comment keyword features can be used to optimize the quantitative frequent items of opinion comments to obtain the current opinion comment frequent items, so as to obtain the encoding vector (quantitative encoding feature) of the topic interest and focus points that is as comprehensive and complete as possible and can represent the relevant feature semantics.

[0073] It can be understood that when topic interest capture is performed through the above topic interest capture network, in an embodiment of the present invention, target comment keyword features that are associated with the quantified frequent items of opinion comments can be determined based on the comment keyword feature components in the topic interest capture network.

[0074] In some exemplary design ideas, for the same topic of interest, different online topic interaction scenarios may make the corresponding text detail representation vectors different. Therefore, in the embodiment of the present invention, different comment keyword feature pools corresponding to different online topic interaction scenarios can be generated. At this time, the implementation method of step 203 can be to obtain the comment keyword feature pool corresponding to the selected online topic interaction scenario, and determine the target comment keyword feature that is related to the quantified frequent item of the opinion comment from the comment keyword feature pool corresponding to the selected online topic interaction scenario.

[0075] For example, comment keyword feature pools corresponding to VR mall interaction scenarios, VR shopping guide interaction scenarios, and digital space business interaction scenarios are generated respectively. If the user session activity records to be analyzed are collected in the VR shopping guide interaction scenarios, then when determining the target comment keyword features, it is necessary to determine the commonality measurement results of the quantified frequent items of opinion comments and the comment keyword features in the comment keyword feature pool corresponding to the VR mall interaction scenario, and then determine the target comment keyword features from the comment keyword feature pool corresponding to the VR mall interaction scenario.

[0076] It can be seen that compared with the traditional solution of using fixed feature spaces in various online topic interaction scenarios to optimize the detail vectors of topic interest points, the introduction of scene features in different online topic interaction scenarios and the target comment keyword features in the same online topic interaction scenario can provide a better reference, further improving the accuracy and rationality of subsequent topic interest capture.

[0077] Step 204: Optimize the opinion review quantified frequent items according to the target review keyword features to obtain the current opinion review frequent items.

[0078] Generally, the reference value of the comment keyword features (target comment keyword features) with the same feature semantics is higher. The vector aggregation component of the embodiment of the present invention can be a vector aggregation component based on the Attention Mechanism, and the frequent items of opinion comment quantification are optimized in combination with the feature local reinforcement processing. Under some exemplary design ideas, a credible list of commonality measurement results can be determined based on the target comment keyword features and the frequent items of opinion comment quantification. The members in the credible list of commonality measurement results are used to mark the importance coefficient of each target comment keyword feature for the frequent items of opinion comment quantification. Then, based on the credible list of commonality measurement results, the target comment keyword features and the frequent items of opinion comment quantification are enhanced and integrated to obtain the current frequent items of opinion comment quantification corresponding to the frequent items of opinion comment quantification. Among them, the ideas for determining the credible list of commonality measurement results include: M = function (F*VT / r1 / 2).

[0079] Among them, F is the frequent item of opinion and comment quantization, the dimension is x*r, x is the number of feature vectors in F (the number of topic interest points in the user session activity record to be analyzed), r is the feature dimension, generally 1024; V is the target comment keyword feature, VT is the transpose of V; M is the credible list of commonality measurement results, the dimension is [x*y], indicating the importance coefficient of each target comment keyword feature to the frequent item of opinion and comment quantization F, x represents the number of frequent items of opinion and comment quantization, y represents the number of comment keyword features in the comment keyword feature component; function is the softmax function. Based on the enhanced integration, the frequent items of opinion and comment quantization for each topic interest point are generated for their own current frequent items of opinion and comment. In order not to lose the original details of the topic interest point, it can be optimized on the basis of the original detail vector (frequent item of opinion and comment quantization): G=M*V+F.

[0080] Among them, G is the frequent item of the current opinion and comment, M is the credible list of commonality measurement results, V is the keyword feature of the target comment, and F is the quantified frequent item of the opinion and comment obtained by mining the dialogue opinion and comment.

[0081] It can be understood that when topic interest capture is performed through the above topic interest capture network, in an embodiment of the present invention, the target comment keyword features and the quantized frequent items of opinion comments can be aggregated according to the vector aggregation component in the topic interest capture component to obtain the current opinion comment frequent items.

[0082] Step 205: Determine the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion comment frequent item.

[0083] It can be understood that when topic interest capture is performed through the above topic interest capture network, in an embodiment of the present invention, the target conversation text quantization encoding corresponding to the user conversation activity record to be analyzed can be determined based on the current opinion comment frequent items according to the vector aggregation component.

[0084] In the embodiment of the present invention, there are multiple ways to determine the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion comment frequent item. One way is to directly use the current opinion comment frequent item as the target conversation text quantization code. Another way is to combine the current opinion comment frequent item with the opinion comment quantization frequent item to obtain the target conversation text quantization code. The current opinion comment frequent item and the opinion comment quantization frequent item are combined by combination processing to obtain the target conversation text quantization code, thereby avoiding the loss of original details in the user conversation activity record to be analyzed. If the dimension of the opinion comment quantization frequent item is x*r, the dimension of the obtained target conversation text quantization code is [x*2r].

[0085] Step 206: Capture topic interests of the user conversation activity records to be analyzed based on the target conversation text quantization coding to obtain topic interest capture information.

[0086] In the case of performing topic interest capture through the above topic interest capture network, in an embodiment of the present invention, topic interest capture can be performed on the user session activity record to be analyzed based on the target session text quantization coding according to the topic interest capture component to obtain topic interest capture information. The topic interest capture information may include the feature semantics of the topic interest focus points included in the user session activity record to be analyzed and the distribution window of the topic interest focus points in the user session activity record to be analyzed.

[0087] In the embodiment of the present invention, the comment keyword features of different feature semantics are counted in advance. The comment keyword features are text detail representation vectors corresponding to the topic interest focus points of different feature semantics, which can represent the topic interest of a feature semantics. When capturing the topic interest of the crawled user session activity record to be analyzed in the selected online topic interaction scenario, the dialogue viewpoint comment mining can be performed on the user session activity record to be analyzed to obtain the viewpoint comment quantified frequent items, and then based on the commonality measurement results of the viewpoint comment quantified frequent items and the comment keyword features in the comment keyword feature pool, the target comment keyword features that are related to the viewpoint comment quantified frequent items are determined from the comment keyword feature pool. The comment keyword feature pool includes comment keyword features of different feature semantics, and the comment keyword features are text detail representation vectors corresponding to the topic interest focus points of different feature semantics. Therefore, the viewpoint comment quantified frequent items can be optimized based on the target comment keyword features to obtain the current viewpoint comment frequent items, and the target conversation text quantified encoding corresponding to the user session activity record to be analyzed can be determined based on the current viewpoint comment frequent items, so as to obtain an encoding vector that is as comprehensive and complete as possible and can represent the topic interest focus points of the relevant feature semantics. In particular, when the selected online topic interaction scenario affects the user conversation activity records to be analyzed, resulting in missing frequent items of the mined opinion comments and low feature recognition, the frequent items of opinion comments are optimized based on the keyword features of the target comments to help capture topic interests and ensure the integrity and feature recognition of the obtained target conversation text quantitative coding. In this way, the topic interest capture of the user conversation activity records to be analyzed based on the target conversation text quantitative coding is more accurate and reasonable, which improves the accuracy and credibility of topic interest capture and provides a more credible analysis basis for the big data recommendation of digital service businesses in related digital service industries.

[0088] In some exemplary design ideas, in order to improve the accuracy and rationality of topic interest capture in non-high-quality online topic interaction scenarios, the text retelling network can be pre-debugged, and then after the user session activity records are collected, the crawled user session activity records are retelled according to the text retelling network to improve the quality of the user session activity records input into the topic interest capture network, thereby improving the accuracy and rationality of topic interest capture. Compared with the idea of ​​improving the accuracy and rationality of topic interest capture based on the text retelling network, the big data recommendation method for cloud digital services provided by the embodiment of the present invention does not require debugging of additional neural networks, reduces resource overhead, and increases the scope of application of the solution.

[0089] In some independent embodiments, the debugging step of the topic interest capture network may include the following related content.

[0090] Step 401: obtaining a user session activity record sample in an online topic interaction scenario sample, wherein the user session activity record sample has a priori feature semantic annotations of the topic interest points included therein.

[0091] Among them, the prior feature semantic annotation can be understood as the training label of the feature semantics, and the examples in this embodiment can all be understood as training samples or debugging samples.

[0092] Step 402: Load the user session activity record sample into a general topic interest capture network.

[0093] Among them, the general topic interest capture network includes a general dialogue opinion comment mining component, a general comment keyword feature component, a general vector aggregation component and a general topic interest capture component. The user conversation activity record samples have prior feature semantic annotations of the topic interest points they include.

[0094] Furthermore, the general network / component / model can be understood as an untrained network / component / model, and can also be understood as an initial network / component / model.

[0095] Step 403: perform conversation opinion and comment mining on the user session activity record sample according to the general conversation opinion and comment mining component to obtain opinion and comment quantized frequent item samples.

[0096] Step 404: According to the general comment keyword feature component, based on the commonality measurement results of the opinion comment quantified frequent item samples and the comment keyword features in the comment keyword feature pool, determine the target comment keyword feature samples that are related to the opinion comment quantified frequent item samples from the general comment keyword feature component.

[0097] Step 405: According to the general vector aggregation component, the target comment keyword feature sample is used to optimize the opinion comment quantized frequent item sample to obtain the current opinion comment frequent item sample.

[0098] Step 406: Determine, according to the general vector aggregation component, a target conversation text quantization encoding sample corresponding to the user conversation activity record sample based on the current opinion comment frequent item sample.

[0099] Step 407: According to the general topic interest capture component, the topic interest of the user conversation activity record to be analyzed is captured based on the target conversation text quantization encoding sample to obtain a topic interest capture information sample.

[0100] Step 408: Debug the general topic interest capture network based on the prior feature semantic annotations and the topic interest capture information samples to obtain the topic interest capture network.

[0101] It can be understood that, in the process of debugging the topic interest capture network, the exemplary implementation of steps 401 to 407 can be combined with the above-mentioned corresponding topic interest capture ideas.

[0102] In the process of debugging the topic interest capture network, a comment keyword feature component can be generated. The idea of ​​generating the comment keyword feature component can be that in the process of debugging the topic interest capture network, according to the dialogue viewpoint comment mining component, the user session activity record sample is mined for dialogue viewpoint comments, and the viewpoint comment quantized frequent item samples of the topic interest focus matched by the prior feature semantic annotation are obtained. The prior feature semantic annotation is used to mark the feature semantics of the topic interest focus included in the user session activity record sample; based on the viewpoint comment quantized frequent item samples of the topic interest focus, the comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component are adjusted.

[0103] During the debugging process of the topic interest capture network, an embodiment of the present invention introduces and generates a comment keyword feature component to compensate for the identification features that suppress topic interest focus points based on the text detail representation vector in the comment keyword feature component, which helps to ensure the stability and anti-interference of the topic interest capture network.

[0104] It can be understood that the idea of ​​adjusting the comment keyword features of the feature semantics corresponding to the topic interest and concern points in the comment keyword feature component based on the quantified frequent item samples of opinion comments on the topic interest and concern points can be to use the quantified frequent item samples of opinion comments on the topic interest and concern points as the comment keyword features of the feature semantics corresponding to the topic interest and concern points in the comment keyword feature component.

[0105] In view of the fact that the comment keyword features of the corresponding feature semantics in the comment keyword feature component need to occupy a certain storage area, the more the number of comment keyword features contained, the larger the storage area occupied, the more resource overhead, and the quality of topic interest capture may be affected. Based on this, under some exemplary design ideas, the number of comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component can be adjusted. Generally speaking, if the number of comment keyword features of the feature semantics corresponding to the topic interest focus = 0 (that is, the comment keyword feature set Vk corresponding to the feature semantics is empty), the viewpoint comment quantization frequent item sample of the topic interest focus is used as the comment keyword feature of the feature semantics corresponding to the topic interest focus; if the number of comment keyword features of the feature semantics corresponding to the topic interest focus ≠ 0 (that is, the comment keyword feature set Vk corresponding to the feature semantics is not empty), the commonality measurement result between the viewpoint comment quantization frequent item sample of the topic interest focus and the comment keyword feature of the feature semantics corresponding to the topic interest focus is determined, and the comment keyword feature of the feature semantics corresponding to the topic interest focus in the comment keyword feature component is adjusted based on the commonality measurement result. Among them, the commonality measurement result between the quantified frequent item samples of the opinions and comments on the topic interest focus and the comment keyword features of the feature semantics corresponding to the topic interest focus can be the cosine similarity.

[0106] For some exemplary embodiments, if the commonality measurement result is less than the commonality measurement result limit (for example, the commonality measurement result limit is set to 0.6), the quantified frequent item samples of the opinions and comments of the topic interest points are directly added to the comment keyword features of the feature semantics corresponding to the topic interest points. Among them, the commonality measurement result limit (similarity threshold) can be flexibly set.

[0107] In order to avoid the situation where the number of comment keyword features in the comment keyword feature component is too large due to the addition of frequent items of quantization of opinion comments, for some exemplary embodiments, the idea of ​​adjusting the comment keyword features of the feature semantics corresponding to the topic interest and focus in the comment keyword feature component based on the commonality measurement result can be that if the commonality measurement result is less than the commonality measurement result limit, the comparison result between the number of comment keyword features of the feature semantics corresponding to the topic interest and focus and the number limit is determined, and then the comment keyword features of the feature semantics corresponding to the topic interest and focus in the comment keyword feature component are adjusted based on the comparison result.

[0108] Generally speaking, if the comparison result indicates that the number of comment keyword features corresponding to the feature semantics of the topic interest focus in the comment keyword feature component reaches the number limit (the number limit can be set to 50), it means that the comment keyword features in the comment keyword feature component are saturated. At this time, the comment keyword feature with the highest commonality measurement result can be replaced with the frequent item of opinion comment quantization. Among them, the number limit can be set flexibly. Based on this idea, the number of comment keyword features can be reduced as much as possible on the premise of obtaining the text detail representation vector of the topic interest focus, thereby reducing the occupancy of the storage area and ensuring the quality of capturing the topic interest.

[0109] In this way, as the topic interest capture network is debugged, the comment keyword feature component containing the text detail representation vector can be obtained, and this method of obtaining it enables the entire comment keyword feature component to be fully debugged (end-to-end debugging) along with the topic interest capture network.

[0110] Exemplarily, the process of adjusting the comment keyword features corresponding to the feature semantics of the topic interest focus in the comment keyword feature component (class center component) may include the following: for each user session activity record sample in a group, all capture windows (also called certified capture windows) can be obtained using the authenticated annotated prior feature semantic annotations, and the viewpoint comment quantified frequent items of the topic interest focus corresponding to the capture window can be obtained based on the reversible layer. Each viewpoint comment quantified frequent item has its own correct prior feature semantic annotation, and the obtained viewpoint comment quantified frequent item is used to adjust the comment keyword features corresponding to its own prior feature semantic annotation.

[0111] Furthermore, the prior feature semantic annotations include prior feature semantic annotation 1 (tag_1), prior feature semantic annotation 2 (tag_2), prior feature semantic annotation 3 (tag_3), ..., the number of comment keyword features of tag_1 has reached the number limit, and since the commonality measurement results of the latest opinion and comment quantization frequent item corresponding to tag_1 and all comment keyword features in the comment keyword features corresponding to tag_1 are relatively low, the most similar comment keyword feature is replaced based on the latest opinion and comment quantization frequent item corresponding to tag_1. The number of comment keyword features of tag_2 does not reach the number limit, and the commonality measurement results of the latest opinion and comment quantization frequent item of tag_2 and all comment keyword features in the comment keyword features of tag_2 are lower than the commonality measurement result limit, so this opinion and comment quantization frequent item is added to the comment keyword features of tag_2. Even if the number of comment keyword features of tag_3 does not reach the number limit, the latest opinion and comment quantization frequent item of tag_3 is very similar to all comment keyword features in the comment keyword features of tag_3, so the opinion and comment quantization frequent item of tag_3 is ignored.

[0112] It is understandable that when adjusting the comment keyword features corresponding to the feature semantics of the topic interest focus in the comment keyword feature component, in order to obtain the current viewpoint comment frequent item G with high feature quality, a target debugging cost indicator can be generated to guide the current viewpoint comment frequent item. For example, the target debugging cost indicator is generated based on the difference between the prior feature semantic annotation and the obtained current viewpoint comment frequent item sample, and the comment keyword features in the comment keyword feature component are adjusted based on the target debugging cost indicator to generate the comment keyword feature component.

[0113] Among them, the target debugging cost indicator can be different categories of loss functions, such as cross entropy loss function, triple loss function, hinge loss function, etc.

[0114] In an embodiment of the present invention, when adjusting the comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component, a target debugging cost indicator for feature selection is configured therefor, so that the text detail representation vector can be flexibly managed and network training debugging can be accurately realized.

[0115] In some independent embodiments, the topic interest capture information is used to guide the big data recommendation of digital service business of the digital service industry corresponding to the selected online topic interaction scenario. Based on this, after the topic interest capture of the user session activity record to be analyzed based on the target session text quantization coding to obtain the topic interest capture information, the method also includes: generating current push demand information for the target user through the topic interest capture information; extracting the push attribute characteristics of each push demand item in the current push demand information and the push attribute characteristics of each push demand item in the historical push demand information respectively; determining the intra-class attribute characteristics of the current push demand information and the intra-class attribute characteristics of the historical push demand information based on the push attribute characteristics of each push demand item in the current push demand information and the push attribute characteristics of each push demand item in the historical push demand information respectively; based on each of the current push demand information Determine the inter-class attribute characteristics of the current push demand information based on the push attribute characteristics of the push demand items and the intra-class attribute characteristics of the historical push demand information, and determine the inter-class attribute characteristics of the historical push demand information based on the push attribute characteristics of each push demand item in the historical push demand information and the intra-class attribute characteristics of the current push demand information; determine the push strategy adaptation characteristics of the current push demand information and the historical push demand information based on the intra-class attribute characteristics of the current push demand information, the inter-class attribute characteristics of the current push demand information, the intra-class attribute characteristics of the historical push demand information and the inter-class attribute characteristics of the historical push demand information; when it is determined based on the push strategy adaptation characteristics that the current push demand information and the historical push demand information meet the push strategy conversion conditions, generate the current push policy for the current push demand information by utilizing the historical push policy of the historical push demand information.

[0116] In some independent embodiments, push attribute features of each push demand item in the current push demand information and push attribute features of each push demand item in the historical push demand information are extracted respectively, including: extracting push attribute features of each push demand item in the current push demand information and push attribute features of each push demand item in the historical push demand information through a first convolutional model.

[0117] In some independent embodiments, after extracting the push attribute characteristics of each push demand item in the current push demand information and the push attribute characteristics of each push demand item in the historical push demand information, the method also includes: downsampling the push attribute characteristics of each push demand item in the current push demand information and the push attribute characteristics of each push demand item in the historical push demand information through the first processing unit of the first convolution model to obtain the first downsampled attribute characteristics of each push demand item in the current push demand information and the first downsampled attribute characteristics of each push demand item in the historical push demand information.

[0118] In some independent embodiments, determining the intra-class attribute characteristics of the current push demand information and the intra-class attribute characteristics of the historical push demand information based on the push attribute characteristics of each push demand item in the current push demand information and the push attribute characteristics of each push demand item in the historical push demand information includes: inputting the push attribute characteristics of each push demand item in the current push demand information and the first down-sampled attribute characteristics of each push demand item in the current push demand information into a second convolutional model to determine the intra-class attribute characteristics of the current push demand information; inputting the push attribute characteristics of each push demand item in the historical push demand information and the first down-sampled attribute characteristics of each push demand item in the historical push demand information into a second convolutional model to determine the intra-class attribute characteristics of the historical push demand information.

[0119] In the embodiment of the present invention, the push attribute features are used to reflect a series of features of the push demand items, such as the push mode, push content, and push time period, while the intra-class attribute features reflect the various attribute features of different push demand information, and the inter-class attribute features reflect the attribute features of the correlation between different push demand information. Based on this, the push strategy adaptation features can be completely and comprehensively determined, and then the push strategy adaptation features can be used to determine whether the current push demand information and the historical push demand information meet the push strategy conversion conditions from the feature similarity level. If so, the historical push strategy of the historical push demand information is used to generate the current push strategy for the current push demand information. In this way, the previous data information can be used as a reference to quickly and accurately customize the push strategy of the current push demand information and reduce unnecessary resource overhead.

[0120] Based on the same or similar technical concepts mentioned above, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and the computer program executes a big data recommendation method applied to cloud digital services when running.

[0121] Based on the same or similar technical concepts mentioned above, an embodiment of the present invention further provides a computer program product, including a computer program or computer executable instructions, which, when executed by a processor, implements a big data recommendation method applied to cloud digital services.

[0122] The embodiment of the present invention provides a big data recommendation method and a big data mining system applied to cloud digital services. The present invention counts the comment keyword features of different feature semantics in advance. The comment keyword features are text detail representation vectors corresponding to the topic interest focus points of different feature semantics, which can represent the topic interest of a feature semantics. When capturing the topic interest of the crawled user session activity record to be analyzed in the selected online topic interaction scenario, the conversation viewpoint comment mining can be performed on the user session activity record to be analyzed to obtain the viewpoint comment quantified frequent items, and then based on the commonality measurement results of the viewpoint comment quantified frequent items and the comment keyword features in the comment keyword feature pool, the target comment keyword features related to the viewpoint comment quantified frequent items are determined from the comment keyword feature pool. The comment keyword feature pool includes comment keyword features of different feature semantics, and the comment keyword features are text detail representation vectors corresponding to the topic interest focus points of different feature semantics. Therefore, the viewpoint comment quantified frequent items can be optimized based on the target comment keyword features to obtain the current viewpoint comment frequent items, and the target conversation text quantified encoding corresponding to the user session activity record to be analyzed is determined based on the current viewpoint comment frequent items, so as to obtain the encoding vector that is as comprehensive and complete as possible and can represent the topic interest focus points of the relevant feature semantics. In particular, when the selected online topic interaction scenario affects the user conversation activity records to be analyzed, resulting in the missing of the mined opinion and comment quantification frequent items and low feature recognition, the opinion and comment quantification frequent items are optimized according to the target comment keyword features to help capture the topic interest, and ensure the integrity and feature recognition of the obtained target conversation text quantification coding. In this way, the topic interest of the user conversation activity records to be analyzed is captured based on the target conversation text quantification coding, and the obtained topic interest capture information is more accurate and reasonable, which improves the accuracy and credibility of the topic interest capture, and provides a more credible analysis basis for the big data recommendation of digital service businesses in the relevant digital service industry. In summary, the embodiments of the present invention improve the technical problem that the prior art is difficult to accurately and reliably realize topic interest capture, and thus improve the technical problem that it is difficult to provide a more credible analysis basis for the big data recommendation of digital service businesses in the relevant digital service industry.

[0123] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0124] The above disclosure is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the relevant contents of the present invention are still within the scope of the present invention.

Claims

1. A big data recommendation method applied to cloud digital business, characterized in that: Applied to a big data mining system, the method comprises: Obtaining the crawled user session activity records to be analyzed in the selected online topic interaction scenario; performing conversation opinion and comment mining on the user session activity records to be analyzed to obtain quantified frequent items of opinion and comment; Based on the commonality measurement results of the frequent quantified items of the opinion comments and the comment keyword features in the comment keyword feature pool, a target comment keyword feature that is related to the frequent quantified items of the opinion comments is determined from the comment keyword feature pool, wherein the comment keyword feature pool includes comment keyword features with different feature semantics, and the comment keyword feature is a text detail representation vector corresponding to the topic interest focus points with different feature semantics; The quantitative frequent items of the opinion comments are optimized according to the keyword features of the target comments to obtain the current opinion comment frequent items; the target conversation text quantitative coding corresponding to the user conversation activity record to be analyzed is determined based on the current opinion comment frequent items; the topic interest of the user conversation activity record to be analyzed is captured based on the target conversation text quantitative coding to obtain topic interest capture information, so as to guide the big data recommendation of the digital service business corresponding to the selected online topic interaction scenario based on the topic interest capture information.

2. The method according to claim 1, characterized in that The method further comprises: Loading the user session activity record to be analyzed into a topic interest capture network, wherein the topic interest capture network includes a conversation viewpoint comment mining component, a comment keyword feature component, a vector aggregation component and a topic interest capture component, wherein the comment keyword feature component is formed by comment keyword features in the comment keyword feature pool; The step of mining the conversation viewpoints and comments of the user session activity record to be analyzed to obtain the viewpoint and comments quantified frequent items includes: mining the conversation viewpoints and comments of the user session activity record to be analyzed according to the conversation viewpoint and comments mining component to obtain the viewpoint and comments quantified frequent items; The method of determining a target comment keyword feature associated with the opinion comment quantification frequent item from the comment keyword feature pool based on the commonality measurement result of the opinion comment quantification frequent item and the comment keyword feature in the comment keyword feature pool comprises: determining the target comment keyword feature associated with the opinion comment quantification frequent item based on the comment keyword feature component; The step of optimizing the viewpoint review quantified frequent item according to the target review keyword feature to obtain the current viewpoint review frequent item includes: aggregating the target review keyword feature and the viewpoint review quantified frequent item according to the vector aggregation component to obtain the current viewpoint review frequent item; The determining the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion and comment frequent items includes: determining the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion and comment frequent items according to the vector aggregation component; The topic interest capturing of the user conversation activity record to be analyzed based on the quantization encoding of the target conversation text to obtain the topic interest capturing information includes: the topic interest capturing of the user conversation activity record to be analyzed based on the quantization encoding of the target conversation text according to the topic interest capturing component to obtain the topic interest capturing information.

3. The method according to claim 2, characterized in that The method further includes generating the comment keyword feature component, wherein generating the comment keyword feature component includes: In the process of debugging the topic interest capture network, the user session activity record sample is mined for dialogue opinion and comment according to the dialogue opinion and comment mining component to obtain the quantized frequent item samples of the opinion and comment of the topic interest focus matched by the prior feature semantic annotation, wherein the prior feature semantic annotation is used to mark the feature semantics of the topic interest focus included in the user session activity record sample; The comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component are adjusted based on the quantified frequent item samples of the opinion comments on the topic interest focus.

4. The method according to claim 3, characterized in that The step of adjusting the comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component based on the frequent item examples of the opinion comments on the topic interest focus includes: If the number of comment keyword features of the characteristic semantics corresponding to the topic interest focus point is 0, then the quantified frequent item sample of the opinion comment of the topic interest focus point is used as the comment keyword feature of the characteristic semantics corresponding to the topic interest focus point; If the number of the comment keyword features of the characteristic semantics corresponding to the topic interest focus point is ≠ 0, then determining the commonality measurement result between the quantized frequent item samples of the opinion comments of the topic interest focus point and the comment keyword features of the characteristic semantics corresponding to the topic interest focus point; Adjusting the comment keyword features of the feature semantics corresponding to the topic interest focus in the comment keyword feature component based on the commonality measurement result; Among them, the adjusting of the comment keyword features of the feature semantics corresponding to the topic interest and focus in the comment keyword feature component based on the commonality measurement result includes: if the commonality measurement result is less than the commonality measurement result limit, adding the quantized frequent item samples of the opinion comments on the topic interest and focus to the comment keyword features of the feature semantics corresponding to the topic interest and focus; or, if the commonality measurement result is less than the commonality measurement result limit, determining the comparison result between the number of comment keyword features of the feature semantics corresponding to the topic interest and focus and the number limit; and adjusting the comment keyword features of the feature semantics corresponding to the topic interest and focus in the comment keyword feature component based on the comparison result.

5. The method according to claim 3, characterized in that The step of generating the comment keyword feature component includes: Generate a target debugging cost indicator based on the difference between the prior feature semantic annotation and the obtained current opinion comment frequent item sample; Adjusting the comment keyword features in the comment keyword feature component based on the target debugging cost indicator to generate the comment keyword feature component; Among them, different online topic interaction scenarios correspond to different comment keyword feature pools, and the target comment keyword feature that is related to the viewpoint comment quantification frequent item is determined from the comment keyword feature pool based on the commonality measurement result of the viewpoint comment quantification frequent item and the comment keyword feature in the comment keyword feature pool, including: obtaining the comment keyword feature pool corresponding to the selected online topic interaction scenario; determining the target comment keyword feature that is related to the viewpoint comment quantification frequent item from the comment keyword feature pool corresponding to the selected online topic interaction scenario; The step of determining the target conversation text quantization code corresponding to the user conversation activity record to be analyzed based on the current opinion and comment frequent item includes: combining the current opinion and comment frequent item with the opinion and comment quantization frequent item to obtain the target conversation text quantization code.

6. The method according to claim 1, characterized in that The step of optimizing the quantified frequent items of the opinion comments according to the keyword features of the target comments to obtain the current frequent items of the opinion comments includes: Calculating a commonality measurement result trust list based on the target comment keyword feature and the opinion comment quantization frequent item, wherein the members in the commonality measurement result trust list are used to mark the importance coefficient of each target comment keyword feature to the opinion comment quantization frequent item; Based on the commonality measurement result credible list, the target comment keyword features and the opinion comment quantification frequent items are enhanced and integrated to obtain the current opinion comment frequent items corresponding to the opinion comment quantification frequent items.

7. The method according to claim 2, characterized in that The method further comprises: Obtaining a user session activity record sample under an online topic interaction scenario sample, wherein the user session activity record sample has a priori feature semantic annotations of the topic interest points included therein; The user session activity record sample is loaded into a general topic interest capture network, wherein the general topic interest capture network includes a general dialogue opinion comment mining component, a general comment keyword feature component, a general vector aggregation component and a general topic interest capture component, and the user session activity record sample has a prior feature semantic annotation of the topic interest focus included therein; Performing conversation opinion and comment mining on the user session activity record sample according to the general conversation opinion and comment mining component to obtain opinion and comment quantized frequent item samples; According to the general comment keyword feature component, based on the commonality measurement results of the opinion comment quantization frequent item sample and the comment keyword features in the comment keyword feature pool, determine the target comment keyword feature sample that is related to the opinion comment quantization frequent item sample from the general comment keyword feature component; According to the general vector aggregation component, the target comment keyword feature sample is used to optimize the opinion comment quantized frequent item sample to obtain the current opinion comment frequent item sample; According to the general vector aggregation component, based on the current opinion review frequent item sample, determine the target conversation text quantization encoding sample corresponding to the user conversation activity record sample; According to the general topic interest capture component, based on the target conversation text quantization coding sample, the topic interest of the user conversation activity record to be analyzed is captured to obtain a topic interest capture information sample; The general topic interest capture network is debugged based on the prior feature semantic annotation and the topic interest capture information sample to obtain the topic interest capture network.

8. A big data mining system, characterized in that: include: A processor, a memory and a network interface; the processor is connected to the memory and the network interface; the network interface is used to provide a data communication function, the memory is used to store program code, and the processor is used to call the program code to execute the big data recommendation method applied to cloud digital business as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is run, the method for recommending big data applied to cloud digital services as described in any one of claims 1 to 7 is executed.

10. A computer program product, characterized in that It includes a computer program or a computer executable instruction, which, when executed by a processor, implements the big data recommendation method applied to cloud digital business as described in any one of claims 1 to 7.

Images