A data processing method and related apparatus

Abstract

Embodiments of the present application disclose a data processing method and related device, for a target product needing data attribution, historical product data related to the target product is obtained, data indicators and corresponding data dimension features in the historical product data are determined, value change conditions of the data indicators and the data dimension features in multiple time windows are determined, and an index dimension ascending / descending sequence with time window granularity is obtained based on the value change conditions. Since the index dimension ascending / descending sequence can reflect the change condition of the correlation between the data indicators and the data dimension features under time change, the data attribution recognition model trained based on the index dimension ascending / descending sequence can recognize data attribution results of the data indicators in different time windows based on target product data. The data attribution recognition model can provide more dynamic data attribution results in the time dimension, and the extraction of data dimension features associated with the data indicators based on the data attribution results is more flexible.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a data processing method and related devices. Background Art

[0002] Data attribution can be used to analyze the relationships between data metrics and various analytical dimensions. This analysis can provide important guidance for digital platform products at different stages, such as achieving user growth and analyzing user retention.

[0003] Related technologies often use rule engines and expert systems to determine data attribution. This requires the introduction of experts and prior knowledge, and the specified rules are difficult to expand and have low flexibility. For complex and changeable operation and maintenance environments, the coverage of the originally specified rules will often decrease significantly after a certain period of time, requiring continuous manual updates of the knowledge base and rule base, which is very costly.

[0004] It can be seen that how to effectively attribute data is a technical problem that needs to be solved urgently. Summary of the Invention

[0005] In order to solve the above technical problems, the present application provides a data processing method and related devices that can adapt to the needs of dynamic operation and maintenance changes, and are more flexible in extracting data dimension features associated with data indicators based on data attribution results.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In one aspect, an embodiment of the present application provides a data processing method, the method comprising:

[0008] Obtain historical product data related to the target product;

[0009] Determining a data indicator corresponding to the historical product data, and a data dimension feature in the historical product data corresponding to the data indicator;

[0010] Determine, based on the numerical changes of the data indicators and the corresponding data dimension features in a plurality of consecutive time windows, the indicator dimension ascending and descending sequences corresponding to the plurality of time windows, wherein, for the i-th time window of the plurality of time windows, the corresponding indicator dimension ascending and descending sequence is used to identify the ascending and descending information of the data indicators and the corresponding data dimension features in the i-th time window compared to the i-1-th time window;

[0011] A data attribution recognition model is trained based on the ascending and descending sequence of the indicator dimension and the actual data attribution label of the data indicator determined based on the historical product data;

[0012] obtain target product data of the target product, the target product data being determined according to the to-be-identified data indicator;

[0013] determine, by the data attribution recognition model, a data attribution result corresponding to the to-be-identified data indicator according to the target product data.

[0014] In another aspect, an embodiment of the present application provides a data processing apparatus, the apparatus comprising an obtaining unit, a determining unit and an identifying unit:

[0015] The obtaining unit is configured to obtain historical product data related to a target product.

[0016] The determining unit is configured to determine a data indicator corresponding to the historical product data, and a data dimension feature corresponding to the data indicator in the historical product data.

[0017] The determining unit is further configured to determine, according to the data indicator and the numerical change of the corresponding data dimension feature in a plurality of continuous time windows, an indicator dimension ascending / descending sequence corresponding to the data indicator in each of the plurality of time windows, the indicator dimension ascending / descending sequence corresponding to an i-th time window in the plurality of time windows being used to identify the ascending / descending information of the data indicator and the corresponding data dimension feature in the i-th time window compared with an (i-1)-th time window.

[0018] The determining unit is further configured to train a data attribution recognition model according to the indicator dimension ascending / descending sequence and an actual data attribution label of the data indicator determined based on the historical product data.

[0019] The obtaining unit is further configured to obtain target product data of the target product, the target product data being determined according to the to-be-identified data indicator.

[0020] The identifying unit is configured to determine, by the data attribution recognition model, a data attribution result corresponding to the to-be-identified data indicator according to the target product data.

[0021] In yet another aspect, an embodiment of the present application provides a computer device, the computer device comprising a processor and a memory:

[0022] The memory is configured to store program code and transmit the program code to the processor.

[0023] The processor is configured to execute the method according to the instructions in the program code.

[0024] In yet another aspect, the embodiments of the present application provide a computer readable storage medium for storing a computer program for executing the method of the above aspect.

[0025] In yet another aspect, the embodiments of the present application provide a computer program product comprising instructions which, when executed on a computer, cause the computer to carry out the method of the above aspect.

[0026] From the above technical solution, for a target product that needs data attribution, historical product data related to the target product is obtained, data indicators and corresponding data dimension characteristics in the historical product data are determined, value changes of the data indicators and the data dimension characteristics in multiple time windows are determined, and an index dimension ascending / descending sequence with time window granularity is obtained based on the value changes. Since the index dimension ascending / descending sequence identifies the value ascending / descending information of the data indicators and the data dimension characteristics in different time windows, it can reflect the change of the correlation between the data indicators and the data dimension characteristics under the change of time. Therefore, according to the index dimension ascending / descending sequence and the actual data attribution label of the data indicators determined based on the historical product data, the data attribution recognition model obtained by training can recognize the data attribution result of the to-be-recognized data indicators in different time windows based on the target product data. Compared with the fixed expert rule, the data attribution recognition model can provide more dynamic data attribution results in the time dimension, can adapt to the needs of dynamic operation and maintenance changes, and can extract data dimension characteristics associated with data indicators based on the data attribution result more flexibly. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0028] Figure 1 A data attribution determination scenario diagram is provided for the embodiments of the present application.

[0029] Figure 2 A method flowchart of a data processing method is provided for the embodiments of the present application.

[0030] Figure 3 A schematic diagram of a Transformer model for determining dimension characteristic correlation is provided for the embodiments of the present application.

[0031] Figure 4 A schematic diagram of a data attribution recognition model is provided for the embodiments of the present application.

[0032] Figure 5 A schematic diagram of a hierarchical topology network provided for an embodiment of the present application;

[0033] Figure 6a A schematic diagram of an index dimension link generation based on a hierarchical topology network provided for an embodiment of the present application;

[0034] Figure 6b A determination scenario schematic diagram of a data attribution result set provided for an embodiment of the present application;

[0035] Figure 7 A device structure diagram of a data processing apparatus provided for an embodiment of the present application;

[0036] Figure 8 A structure diagram of a terminal device provided for an embodiment of the present application;

[0037] Figure 9 A structure diagram of a server provided for an embodiment of the present application. DETAILED DESCRIPTION

[0038] The embodiments of the present application are described below with reference to the accompanying drawings.

[0039] The core and purpose of attribution analysis is to quickly locate the correlation of business indicators. The methods for attribution analysis in related technologies mainly include: data attribution based on rule engine and expert system, data attribution based on reasoning tree, and data attribution based on knowledge graph. The attribution analysis based on rule engine and expert system is usually suitable for early system operation and maintenance, and needs a large number of manually specified expert system rules for summary and deduction. For example, the Drools rule engine is used to realize, the deduction rules are continuously enriched and improved, so that the system has data attribution capability. The data attribution based on reasoning tree refines the experience of problem troubleshooting process into a binary decision tree, classifies and groups the alarm information according to the time slicing algorithm, and finally outputs the grouped alarm information to the decision tree for automatic reasoning to output the reasoning result. The data attribution based on knowledge graph is to generate the knowledge graph of the abnormal event by using the system to collect information, which often needs to combine related algorithm models of the knowledge graph. First, the event is taken as the starting point to query the index information related to the abnormal event, the business flow information at the abnormal time point is obtained, the business flow log and real-time tree log associated with the corresponding business flow number are queried, the existing evidence at that time is obtained, and then all the data is written into the graph database to generate the knowledge graph. In the root cause positioning stage, the abnormal cause is extracted from the knowledge graph by applying the deduction model based on the abnormal event knowledge graph.

[0040] The data attribution based on the rule engine and the expert system needs to introduce experts and priori knowledge, and the specified rules are not easy to extend, have low flexibility, and for complex and changeable operation and maintenance environment, the rule coverage will be significantly reduced after a certain time, and the knowledge base and rule base need to be manually updated constantly, which has high cost. The data attribution based on the reasoning tree has the advantage of strong explainability of the decision tree, but needs to construct the reasoning tree according to the artificial fault positioning judgment logic and clarify the correlation between alarms, which has relatively large workload. The data attribution based on the knowledge graph often needs to collect sufficient information to construct the knowledge graph, and the influence on the analysis result is large for the data fault condition, and in addition, if the graph model and other technologies are involved, the construction of the graph network has high complexity with the increase of the network nodes, and the training process also consumes more time, which significantly affects the industrial operation and maintenance diagnosis time and efficiency.

[0041] Therefore, an embodiment of the present application provides a data attribution determination method and related device, which can adapt to the demand of dynamic operation and maintenance change, and is more flexible in extracting data dimension features associated with data indicators based on data attribution results.

[0042] The data processing method provided in the embodiments of the present application can be implemented by a computer device, which can be a terminal device or a server. The server can be a physical server, a server cluster composed of multiple physical servers or a distributed system, or a cloud server providing cloud computing services. The terminal device includes but is not limited to a mobile phone, a computer, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal and the like. The terminal device and the server can be directly or indirectly connected through wired or wireless communication, which is not limited in the present application.

[0043] The embodiments of the present application relate to artificial intelligence. Artificial intelligence (AI) is the use of digital computers or computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which tries to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to design and implement principles and methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0044] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software level technology. Artificial intelligence basic technology generally includes, such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, automatic driving, intelligent transportation and several other major directions.

[0045] Machine learning (ML) is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent. Its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rule-based learning.

[0046] For example, based on machine learning, the data attribution recognition model is trained by the index dimension ascending / descending sequence and the actual data attribution label of the data index determined based on the historical product data. The data attribution recognition model can be various models as follows, or involve interaction with the following models.

[0047] Convolutional Neural Networks (CNN) is a class of feedforward neural networks containing convolutional computation and having a deep structure, which is one of the representative algorithms of deep learning. Convolutional neural networks have representation learning capabilities and can perform translation-invariant classification on input information according to their hierarchical structure, so they are also called translation-invariant artificial neural networks.

[0048] Recurrent Neural Network (RNN) is a type of recurrent neural network that takes sequence data as input and performs recursion in the evolution direction of the sequence, with all nodes (recurrent units) connected in a chain.

[0049] Transformer is a new model proposed in the paper Attention Is All You Need by Google, which takes the attention mechanism (Attention) to the extreme. It discards the CNN and RNN used in previous deep learning tasks and is widely used in natural language processing, such as machine translation, question answering systems, text summarization and speech recognition.

[0050] Embodiments of the present application can relate to cloud technology and cloud gaming. Cloud technology refers to a series of resources such as hardware, software, network, etc. being unified in a wide area network or a local area network to realize data calculation, storage, processing and sharing. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on cloud computing business model application, which can form a resource pool, and be used on demand, flexibly and conveniently. Cloud computing technology will become an important support. The background service of a technical network system needs a large amount of computing and storage resources, such as video websites, picture websites and more portals. With the high development and application of the Internet industry, in the future, every item may have its own identification mark and needs to be transmitted to the background system for logical processing. Different levels of data will be processed separately, and various industry data need strong system support, which can only be realized through cloud computing. Cloud gaming is an online gaming technology based on cloud computing technology. Cloud gaming technology enables light devices with relatively limited graphics processing and data operation capabilities to run high-quality games. Based on cloud gaming, game on demand or game live streaming can be performed.

[0051] The method provided by the embodiments of the present application can also relate to a blockchain, that is, the method provided by the embodiments of the present application can be implemented based on a blockchain, or the data involved in the method provided by the embodiments of the present application can be stored based on a blockchain, or the execution subject of the method provided by the embodiments of the present application can be located in a blockchain. The blockchain is a new application mode of distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm and other computer technologies. Blockchain, in essence, is a decentralized database, which is a series of data blocks associated using cryptographic methods, each data block containing information about a batch of network transactions, used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer and an application service layer.

[0052] The blockchain underlying platform can include user management, basic services, smart contracts, and operation monitoring processing modules. Among them, the user management module is responsible for the identity information management of all blockchain participants, including maintaining public and private key generation (account management), key management, and user real identity and blockchain address correspondence maintenance (permission management), etc., and under authorization, supervises and audits the transaction of certain real identities, provides risk control rule configuration (risk audit); the basic service module is deployed on all blockchain node devices to verify the validity of business requests, and after consensus, the valid request is recorded on the storage. For a new business request, the basic service first performs interface adaptation analysis and authentication processing (interface adaptation), then encrypts the business information through a consensus algorithm (consensus management), and after encryption, the complete and consistent transmission is transmitted to the shared ledger (network communication), and recorded and stored; the smart contract module is responsible for contract registration and issuance, contract triggering and contract execution. Developers can define contract logic through a certain programming language, publish it to the blockchain (contract registration), trigger execution according to the logic of the contract terms, complete the contract logic, and also provide contract upgrade and cancellation functions; the operation monitoring module is mainly responsible for the deployment, configuration modification, contract setting, cloud adaptation of the product release process, and the real-time state visualization output of the product running, such as alarm, monitoring network condition, monitoring node device health status, etc.

[0053] The platform product service layer provides basic capabilities and implementation frameworks for typical applications. Developers can add business features based on these basic capabilities to complete the blockchain implementation of business logic. The application service layer provides application services based on the blockchain solution for business participants to use.

[0054] It can be understood that in the specific embodiments of the present application, the historical product data may involve user information and other related data. When the above embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0055] As shown in the data attribution determination scenario Figure 1 The server 100 is taken as an example to illustrate the foregoing computer device.

[0056] Since there is no correlation between a data index and any one of the corresponding multiple data dimension features in the past (i.e., having data attribution), it does not mean that there is no correlation in the future. This correlation may be adjusted as the system changes and time passes.

[0057] For example, the data indicator can be the number of daily active users (DAU), and it is often necessary to analyze the abnormal rise and fall of DAU to find the root cause, that is, the data dimension feature causing the abnormal rise and fall, and take measures to avoid and guide the possible bad development direction. In such a scenario, the automation of abnormal change attribution is particularly important.

[0058] To this end, before performing data attribution analysis on the target object, the embodiment of the application can use the historical product data related to the target object to enable the server 100 to learn the knowledge of identifying the variable correlation between the data indicator and the data dimension feature.

[0059] The server 100 determines the numerical change of the data indicator and the data dimension feature in the plurality of time windows according to the data indicator and the corresponding data dimension feature in the historical product data, and obtains the indicator dimension rise and fall sequence with time window granularity based on the numerical change. Since the indicator dimension rise and fall sequence identifies the numerical rise and fall information of the data indicator and the data dimension feature in different time windows, it can reflect the change of the correlation between the data indicator and the data dimension feature under the change of time.

[0060] Since the data dimension features causing the rise and fall of the data indicator in different time windows have been determined for different data indicators in the historical product data, the server 100 can determine these data dimension features as actual data attribution labels corresponding to the data indicator, and use the actual data attribution labels as a guide for model training direction in the supervised training process of the data attribution recognition model according to the indicator dimension rise and fall sequence and the actual data attribution labels.

[0061] The data attribution recognition model obtained by training can recognize the data attribution result of the data indicator to be recognized in different time windows based on the target product data. Compared with the fixed expert rule, the data attribution recognition model can provide more dynamic data attribution results in the time dimension, can provide different data dimension features causing the abnormal change of the same data indicator in different time windows, is more suitable for the needs of dynamic operation and maintenance changes, and is more flexible in extracting the data dimension features associated with the data indicator based on the data attribution result.

[0062] Figure 2 The method flowchart of the data attribution determination method provided by the embodiment of the application is described in the embodiment by taking the server as the foregoing computer device, and the method comprises:

[0063] S201: Obtain historical product data related to a target product.

[0064] The target product is a product that needs to be subjected to data attribution analysis. The product can be an electronic product such as an application or the like, or a physical product such as a home appliance or a daily necessity sold by a platform.

[0065] The historical product data can be product data generated by the target product, or product data generated by other products that are related to the target product. The historical product data identifies various types of data generated by the corresponding product on the Internet, which can include user-related interaction data, or data unrelated to the user.

[0066] S202: Determine the data index corresponding to the historical product data, and the data dimension feature corresponding to the data index in the historical product data.

[0067] Through the historical product data, the data index that needs to be subjected to data attribution analysis can be determined. The data index can be determined according to the operation and maintenance requirements of the target product, such as the DAU described above. The data index determined for different target products can have the same part, or can include different parts.

[0068] For the determined data index, the data dimension feature related to the data index can be obtained according to the relationship between the data. For example, for the data index DAU, the data dimension feature constructed includes: new and old users, user channels, user age, user gender, startup mode, user preferences, place of residence, city level, and the like.

[0069] It should be noted that the historical product data includes various types of data. In order to improve the accuracy of the data attribution recognition model trained subsequently, in one possible implementation manner, the historical product data can be first distinguished based on the data type.

[0070] According to the feature type of the data dimension feature, it can be divided into a continuous type with continuous values (such as the number of users, user age, etc.) and a discrete type with discrete values (such as user gender, startup mode, new and old users, etc.). For the data dimension feature of the continuous type, it can be further divided into a time series type and a non-time series type. The value of the data dimension feature of the time series type will appear value rise and fall with time migration, such as the number of users, index rise and fall information, memory size change, etc. The non-time series type is independent of the product cycle, such as user age. It should be noted that in this application, all other types except the time series type, such as the non-time series type in the continuous type and the discrete type, are collectively referred to as the non-time series type.

[0071] For the data dimension feature of the time series type, the method further includes:

[0072] determine a feature type of a data dimension feature corresponding to the data indicator, the feature type including a time series type of feature numerical values changing over time, and a non-time series type;

[0073] If it is determined that the data dimension feature includes a data dimension feature of the time series type, perform S203 for the data dimension feature of the time series type.

[0074] That is, since the data dimension feature of the time series type has the feature of changing over time relative to the data dimension feature of the non-time series type, the indicator dimension rise and fall sequence determined for the data dimension feature of the time series type is more targeted and can better reflect the numerical value change trend of the data dimension feature of the time series type.

[0075] S203: Determine an indicator dimension rise and fall sequence corresponding to the data indicator and the corresponding data dimension feature changing over time in a plurality of continuous time windows.

[0076] The change in a time window can be determined based on the numerical values of the data indicator and the data dimension feature in the previous time window. For example, the ith time window and the (i-1)th time window are adjacent time windows in the plurality of continuous time windows, and the time defined by the (i-1)th time window is before the time defined by the ith time window. That is, for the ith time window, the corresponding indicator dimension rise and fall sequence is used to identify the rise and fall information of the data indicator and the corresponding data dimension feature in the ith time window compared to the (i-1)th time window.

[0077] The time lengths identified by the plurality of time windows are the same, and the specific time length can be determined based on the data attribution analysis accuracy, or can be determined based on the collection granularity of historical product data. For example, the time length identified by a time window can be one hour, one day, one week, etc.

[0078] Optionally, the expression form of the indicator dimension rise and fall sequence can be as shown in the following table:

[0079] Table 1

[0080]

[0081]

[0082] Wherein, A index is a data index, and B dimension-F dimension can be five data dimension features corresponding to the A index. In Table 1, the time length of the time window identifier is 1 hour, and 24 rows are displayed for the ascending and descending sequence of the index dimension corresponding to 24 time windows in a natural day. For example, for the ascending and descending sequence of the index dimension of the time window "8-9", compared with the values of the A index and the B dimension-F dimension in the time window "7-8", the value of the A index increases, and the values of the five data dimension features of the B dimension-F dimension increase, increase, decrease, increase and decrease, respectively.

[0083] S204: Training a data attribution recognition model according to the index dimension ascending and descending sequence and the actual data attribution label of the data index determined based on the historical product data.

[0084] As described above, the actual data attribution label of the data index in different time windows can be determined based on the historical product data, which identifies which data dimension feature causes the abnormal fluctuation (abnormal rise or fall) of the data index in the corresponding time window. Therefore, the index dimension ascending and descending sequence and the actual data attribution label can be used as training samples to effectively train the recognition model, and the data attribution recognition model is obtained.

[0085] S205: Obtaining target product data of the target product.

[0086] In order to determine the data attribution result corresponding to the to-be-recognized data index, the target product data corresponding to the target product can be obtained according to the to-be-recognized data index, that is, the target product data is determined according to the to-be-recognized data index, which includes the data dimension features related to the to-be-recognized data index.

[0087] S206: Determining the data attribution result corresponding to the to-be-recognized data index according to the target product data through the data attribution recognition model.

[0088] The data attribution model can input the target product data based on the time window in which the to-be-recognized data index fluctuates into the data attribution model, that is, based on the recognition knowledge learned by S201-S204, the data attribution result corresponding to the to-be-recognized data index can be determined based on the data attribution model.

[0089] Therefore, for a target product that needs to attribute data, historical product data related to the target product is obtained, data indicators and corresponding data dimension features in the historical product data are determined, value changes of the data indicators and the data dimension features in a plurality of time windows are determined, and an index dimension ascending / descending sequence with time window granularity is obtained based on the value changes. Since the index dimension ascending / descending sequence identifies the value ascending / descending information of the data indicators and the data dimension features in different time windows, the change of the correlation between the data indicators and the data dimension features under the time change can be reflected. Therefore, according to the index dimension ascending / descending sequence and the actual data attribution label of the data indicators determined based on the historical product data, the data attribution recognition model trained can recognize the data attribution result of the data indicators to be recognized in different time windows based on the target product data. Compared with the fixed expert rule, the data attribution recognition model can provide more dynamic data attribution results in the time dimension, can adapt to the needs of dynamic operation and maintenance changes, and can extract data dimension features associated with the data indicators based on the data attribution result more flexibly.

[0090] In order to better enable the data attribution recognition model to learn the recognition knowledge of the reason for the change of the data indicators in time sequence, in one possible implementation manner, S204 includes:

[0091] S2041: dividing the plurality of time windows into a plurality of investigation periods based on an investigation length, and the investigation period includes N time windows.

[0092] S2042: determining a longest prefix sequence of the data indicators in the kth time window in at least two investigation periods according to the index dimension ascending / descending sequence corresponding to the kth time window in each investigation period and according to a minimum support threshold, wherein the data dimension features included in the longest prefix sequence appear the same ascending / descending information in at least two investigation periods, and the number of times satisfies the minimum support threshold, and k is an element of N.

[0093] S2043: training a data attribution recognition model based on the actual data attribution label according to the longest prefix sequence.

[0094] The investigation period can be determined based on the actual characteristics of the target product and the processing capacity of the application scenario. Since the target product corresponds to a user, the investigation period is generally in natural days, for example, one day, two days, one week, one month, etc. According to the actual time length of the investigation period and the division granularity of the time window, the number N of time windows included in one investigation period can be determined. For example, when the investigation period is one day and the time window identifies a time length of 1 hour, N is 24.

[0095] For the time series similar data dimension feature, the index dimension ascending / descending sequence used for identifying the index-dimension implies the potential causal relationship between the index-dimension, thus it is necessary to mine the potential causal relationship of the index dimension ascending / descending sequence for the index-dimension data attribution analysis. The embodiments of the present application mine the longest prefix sequence as the potential causal relationship based on the prefixspan algorithm, and mine the frequent sequence pattern (i.e. the longest prefix sequence) of each length of each index-dimension value in each time interval in the time range which satisfies the minimum support threshold. Meanwhile, the multi-minimum support strategy is used, and the calculation method of the minimum support is shown in formula (1).

[0096] min_sup=a×n (1)

[0097] Wherein n is the number of observation periods of the index-dimension data time range, for example, it can be the number of days, and a is the minimum support rate, and the minimum support rate parameter is adjusted according to the number of sample sets.

[0098] The prefixspan algorithm is the Prefix-Projected Pattern Growth algorithm, and the frequent sequence pattern can be obtained based on the algorithm, and the PrefixSpan algorithm is also called the prefix projected pattern mining. The prefix in the PrefixSpan algorithm is the subsequence of the front part of the sequence data, for example, for the sequence data B=<a(abc)(ac)d(cf)>, and A=<a(abc)a>, then A is the prefix of B. Of course, there are more than one prefix of B, for example, , <aa><a(ab)> are also prefixes of B. The projection of a prefix is the suffix, for a prefix, the sequence of the remaining subsequence after the prefix is the suffix. In PrefixSpan algorithm, the combination of all suffixes corresponding to the same prefix is the projection database of the prefix.

[0099] The goal of PrefixSpan algorithm is to mine frequent sequences that satisfy the minimum support. It starts from the prefix of length 1 to mine sequence patterns, searches the corresponding projection database to get the frequent sequences of the prefix of length 1, and then recursively mines the frequent sequences of the prefix of length 2, and so on until it cannot mine the frequent sequences of the prefix of longer length. Take the support threshold of 50% as an example. The prefixes of length 1 include< / aa> , , <c> , <d> , <e> , <f> , <g>, which needs to search the corresponding frequent sequences of each prefix recursively.

[0100] The specific operation steps of the algorithm are as follows:

[0101] 1. Find the index-dimension time sequence ascending-descending sequence prefix of unit length 1 and the corresponding projection data set;

[0102] 2. Statistics index-dimension time sequence ascending-descending sequence prefix frequency and add the prefix with support higher than the minimum support threshold to the data set to obtain the frequent one-set sequence mode;

[0103] 3. Recursively mine all prefixes with length i and meeting the minimum support requirement:

[0104] 1) Mine the projection data set of the prefix, and return recursively if the projection data is an empty set;

[0105] 2) Statistics the minimum support of each item in the corresponding projection data set, combine each single item meeting the support with the current prefix to obtain a new prefix, and recursively return if it does not meet the support requirement;

[0106] 3) Let i=i+1, and the prefix is each new prefix after combining the single item, and execute step 3 recursively;

[0107] 4. Return all frequent sequence modes of each route in the index-dimension time sequence ascending-descending sequence sample set.

[0108] The above is the principle description of the Prefixspan algorithm, and the following illustrates the specific mining method of the index-dimension time sequence ascending-descending sequence mode.

[0109] Based on the index-dimension ascending-descending sequence generated by S203, which contains the identifier of the corresponding value change, such as "A increase" for A index increase and "B increase" for B dimension increase, for the same position time window of adjacent observation periods, such as one day observation period and the k=9 time window, the result is as follows:

[0110] Table 2

[0111] Date (Day) Time Period Indicator Dimension Ascending Sequence 20200701 8-9 AM A inc - B inc - C inc - D dec - E inc - F dec 20200702 8-9 AM A dec - B inc - C inc - D dec - E dec - F dec

[0112] Assuming that the minimum support threshold is set to 0.5, the frequency of all identifiers is counted to obtain Table 3.

[0113] Table 3

[0114] Indicator Change A inc A dec B inc C inc D dec E inc E dec F dec Frequency 1 1 2 2 2 1 1 2

[0115] The one-item prefix meeting the threshold and its corresponding suffix are shown in Table 4.

[0116] Table 4

[0117]

[0118] Further, the binomial prefix and the corresponding suffix satisfying the minimum support threshold are shown in Table 5.

[0119] Table 5

[0120]

[0121]

[0122] Further, the trinomial prefix and the corresponding suffix satisfying the minimum support threshold are shown in Table 6.

[0123] Table 6

[0124]

[0125] Further, the tetranomial prefix and the corresponding suffix satisfying the minimum support threshold are shown in Table 7.

[0126] Table 7

[0127] Four Prefixes Corresponding Suffixes B inc - C inc - D dec - F dec

[0128] The mined longest prefix sequence is taken as a frequent sequence pattern of the index-dimension time series ascending-descending sequence, and the pattern is updated in real time with data in a historical time range, and the index-dimension change sequence pattern is mined in real time.

[0129] Similarly, the N longest prefix sequences for the two observation periods can also be determined based on the above manner, and more abundant longest prefix sequences can be obtained based on combinations of different observation periods, so as to play a multi-dimensional effect of describing the change of the data index in time series, and further adapt to the demand of dynamic operation and maintenance change.

[0130] When multiple data dimension features change simultaneously due to the same reason, or different data dimension features appear negatively correlated, the effect of data attribution analysis will be affected. Therefore, the embodiments of the present application provide the correlation degree of the dimension features at different time points through the longest prefix sequence, to obtain more comprehensive information and reduce the influence of the foregoing problems.

[0131] In a possible implementation manner, the method further includes: determining, by the longest prefix sequence corresponding to the data index in the observation period respectively, the correlation degree of the dimension features of the data index between the N time windows based on a multi-head attention model.

[0132] The multi-head attention project can be a Transformer model, which can obtain global information compared to a CNN, and which improves the slow training of an RNN by using a self-attention mechanism to achieve fast parallelism.

[0133] The structure of the Transformer model is described in detail in< / g> < / f> < / e> < / d> < / c> Figure 3 The Multi-Head SelfAttention structure is composed of multiple layers of Attention structures that are completely the same but have different weight matrices, which prevents the model from only paying attention to part of the features of the model. Through the design of Multi-Head, each Head pays attention to different features, and the corresponding model pays attention to more features as a whole, learns the correlation between sequence features at different times, and enables the model to learn different information from different angles and then fuse the information to obtain a better learning result.

[0134] Based on this embodiment, the manner of training the data attribution recognition model in S204 can be:

[0135] According to the dimension feature correlation and the longest prefix sequence, a data attribution recognition model is trained based on the actual data attribution label.

[0136] As shown in FIG. 1, a schematic diagram of a data attribution recognition model in the present application is shown. The data attribution recognition model takes the features in the longest prefix sequence and the dimension feature correlation as input. The features of the longest prefix sequence are input into a Transformer for feature extraction, and the extracted features of the dimension feature correlation and the features of the longest prefix sequence are transmitted to a Gate Recurrent Unit (GRU) layer for fusion. GRU is a kind of recurrent neural network, which has fewer parameters than LSTM and can well process sequence information. The GRU layer can also be omitted and replaced by multiple concatenation layers of feedforward neural network layers, which can also effectively process and fuse features. Figure 4 Next, the fused features are input into a feedforward neural network. Whether the root cause exists is regarded as a probability problem, and a Sigmoid function is used as an output layer to output the probability of whether each data dimension feature is the data attribution of the data indicator. The Sigmoid function is often used as an activation function of a neural network due to its monotonicity and inverse function monotonicity, and can map a variable to between 0 and 1.

[0137] The embodiments of the present application do not limit the extraction method of the features of the dimension feature correlation and the features of the longest prefix sequence. For example, One-Hot encoding can be used to construct the features of the longest prefix sequence, and the features of the dimension feature correlation can be extracted based on convolution operation or attention operation. The training method of the neural network can refer to related technologies, which will not be described here. The training process can use a standard cross-entropy loss function, which can be represented by the following formula:

[0138]

[0139] ​

[0140] Wherein, N, x, y, p(x) represent the number of samples for neural network training, neural network input, true value and neural network predicted value respectively.

[0141] Wherein, the GRU layer is for deep feature extraction, and the GRU layer can also be omitted and replaced by multiple splicing several layers of feedforward neural network layers, which can also effectively process and fuse features. In the embodiment of the application, the network model is constructed with the attribution probability as the output, and the final output data attribution correlation probability is output. If the probability exceeds a certain threshold, it can be considered that the input data dimension has a data attribution relationship compared with the corresponding data index. Figure 4

[0142] As described above, based on the feature type, the data dimension feature can be divided into time series type and non-time series type. Whether the data dimension feature of the time series type has a data attribution relationship with the data index can be determined by the data attribution identification model described above. For the data dimension feature of the non-time series type, the application embodiment also provides a possible implementation manner to determine whether it has a data attribution relationship with the data index.

[0143] S11: If it is determined that the data dimension feature includes a target dimension feature of the non-time series type, a hierarchical topology network between the data index and the target dimension feature is generated according to the parent-child relationship of the target dimension feature in the data dimension. The dimension node in the hierarchical topology network is used to identify the target dimension feature.

[0144] S12: According to the JS divergence and the dimension contribution degree of the dimension node, the dimension nodes whose JS divergence and dimension contribution degree satisfy the preset condition are determined as candidate nodes; wherein the JS divergence is used to identify the probability distribution correlation between the dimension node and the parent node, and the dimension contribution degree is used to identify the contribution degree of the child node of the dimension node to the dimension node.

[0145] S13: According to the candidate nodes in the hierarchical topology network based on the parent-child relationship, an index dimension link constituted with the data index is determined, and the index dimension link is used to identify the data dimension feature having a data attribution relationship with the data index.

[0146] For example Figure 5 As shown in FIG. 8, it is a hierarchical topology network schematic diagram when the data index is DAU. The hierarchical topology network includes multiple levels. The path step number of the dimension node in the same level to the corresponding node of the data index is the same.

[0147] For example Figure 5 ​The path step number of user age and user gender to the root node "number of retained users" is 1, which belongs to the same level of dimension node. The path step number of login method and user channel to the root node "number of retained users" is 3, which also belongs to the same level of dimension node. For the dimension node "login method", its child nodes are PC login, APP login and applet login. Its parent node is the dimension node "male" indicating a user gender.

[0148] The index dimension link is a link formed by the parent-child relationship of the hierarchical topology network and the dimension nodes as candidate nodes. For example, from the number of retained users, through user gender, male, login method, to PC login (assuming all are candidate nodes), it is an index dimension link.

[0149] When the number of retained users changes abnormally (such as a sudden increase or a sudden decrease), the abnormal groups of each dimension node (such as user gender and user age) are identified in the first level. Taking user gender as an example, the JS divergence of the dimension user gender is calculated to determine whether it meets the JS threshold. If it meets the threshold, the dimension user gender is added to the candidate dimension set. JS divergence can measure the difference between two distributions. Combined with JS divergence and contribution, the abnormal index scenario can be better restored, so as to identify the data indicators and data dimension features that have attribution relationship with the preset cutting method.

[0150] KL divergence is a non-symmetrical measure of the difference between two probability distributions P and Q, and is calculated as follows:

[0151]

[0152] As an index for measuring the difference between two distributions, the closer p and q are, the smaller the KL divergence is, and vice versa.

[0153] JS divergence is a symmetrical transformation based on KL divergence, which solves the non-symmetrical problem of KL divergence, and is calculated as follows:

[0154]

[0155] The value range of JS divergence is [0, 1], and P(x) and Q(x) are the probability distribution of the parent node of the dimension node and the probability distribution of the dimension node, respectively. The distribution of the two probability distributions is 0, and the distribution is 1. JS divergence is symmetrical, that is, JS(P||Q) = JS(Q||P).

[0156] Assuming that the JS divergence of the target data dimension feature "user gender" meets the preset JS threshold (preset condition), the dimension user gender is added to the candidate dimension set, and then it is judged whether the dimension corresponding dimension item contribution degree meets the contribution degree threshold, for example: the dimension contribution degree is calculated in combination with the child nodes male and female as the "user gender" dimension item, and the definition of the dimension contribution degree is:

[0157]

[0158] The dimension contribution degree is a criterion including: the dimension contribution degree of a certain dimension item exceeds the contribution degree threshold 1 (preset condition), and the sum of the contribution degrees of the dimension items exceeding the threshold 1 exceeds the contribution degree threshold 2. When a certain dimension item meets the above two conditions, the dimension item is retained as a candidate node and serves as a new big board to continue to divide the dimensions that can identify abnormal groups and construct the index dimension link of the attribution-related dimensions.

[0159] It should be noted that if a dimension node is not determined as a candidate node, the child nodes of the dimension node will not be determined as candidate nodes based on the JS divergence and the dimension contribution value, and if a dimension node is determined as a candidate node, the child nodes of the dimension node will be determined as candidate nodes based on the JS divergence and the dimension contribution value.

[0160] In a possible implementation, the determination of the candidate nodes is described by taking the jth level and the j+1th level in the plurality of levels as an example, S12 includes:

[0161] S21: respectively determining the JS divergence and the dimension contribution degree corresponding to the dimension node at the jth level;

[0162] S22: determining the candidate node at the jth level according to the preset condition;

[0163] S23: determining the candidate node in the j+1th level according to the preset condition based on the child nodes of the candidate node at the jth level in the j+1th level.

[0164] As Figure 6a As shown, after generating a hierarchical topological network between data indicators and target dimensional features according to the target dimensional features, the corresponding JS divergence and dimension contribution are calculated for the dimension nodes in the j-th level of the hierarchical topological network. If the JS divergence and dimension contribution meet the preset conditions, the dimension node is determined as a candidate node, and then the child node of the candidate node of the j-th level in the j+1-th level is determined as the screening object, and the j+1-th level is taken as the j-th level. The JS divergence and dimension contribution of the dimension nodes as the screening objects are traversed, and the candidate nodes are continued to be determined based on the preset conditions until the determination of the candidate nodes of each level of the hierarchical topological network is completed.

[0165] Among them, Figure 5 For example, if the jth level includes user gender and user age, after determining the candidate node from the user gender and user age, it is further determined whether the dimension node of the candidate node in the j+1th level can continue to be a candidate node.

[0166] By processing the aforementioned non-time-series data dimension features, we can obtain an indicator dimension link. When determining the data attribution results for the data indicator to be identified, matching the indicator dimension link can directly determine whether the non-time-series data dimension features of the data indicator to be identified contain the cause of the change in the data indicator to be identified.

[0167] Therefore, in a possible implementation, after obtaining the target product data of the target product in S205, the method further includes:

[0168] S31: Determine a first to-be-identified data dimension feature of a non-time-series type in the target product data;

[0169] S32: Determine, based on the indicator dimension link, a first data attribution result corresponding to the indicator of the data to be identified in the first dimensional feature of the data to be identified.

[0170] That is, by identifying the feature type of the target product data determined by the data indicator to be identified, the non-time-series data dimension feature included therein can be determined and recorded as the first data dimension feature to be identified.

[0171] By matching with the indicator dimension link determined based on S13, if the indicator dimension link can be matched, the first data attribution result related to the non-time series type can be determined based on the matching result. The first data attribution result identifies the common data dimension features of the non-time series type that cause the data indicator to be identified to change.

[0172] Thus, by the above embodiments, the application establishes a data attribution recognition manner for data dimension features of different feature types. When the target product data obtained for the to-be-recognized data indicator includes both the first to-be-recognized data dimension feature belonging to the non-time-series type and the second to-be-recognized data dimension feature belonging to the time-series type, the first data attribution result corresponding to the first to-be-recognized data dimension feature can be determined through the indicator dimension link, and the second data attribution result corresponding to the second to-be-recognized data dimension feature can be determined through the data attribution recognition model.

[0173] Therefore, in a possible implementation, S206 includes:

[0174] determining the second to-be-recognized data dimension feature belonging to the time-series type in the target product data;

[0175] determining, through the data attribution recognition model, a second data attribution result corresponding to the to-be-recognized data indicator according to the second to-be-recognized data dimension feature.

[0176] Correspondingly, after the first data attribution result and the second data attribution result are integrated, a complete data attribution result, a data attribution result set, for the to-be-recognized data indicator can be obtained.

[0177] As shown in FIG. 1, the data indicator and the data dimension feature corresponding to the data indicator are determined for the historical product data. Then, the data dimension features belonging to the time-series type and the data dimension features belonging to the non-time-series type are determined from the data dimension features based on the feature types of the data dimension features. Figure 6b

[0178] For the data dimension feature belonging to the time-series type, the indicator dimension ascending / descending sequence is extracted based on the time window, and the longest prefix sequence is mined. The dimension feature correlation degree is determined according to the longest prefix sequence, so that the data attribution recognition model can be trained according to the dimension feature correlation degree and the longest prefix sequence.

[0179] For the data dimension feature belonging to the non-time-series type, the indicator dimension link is determined from the hierarchical topology network through the JS divergence and the dimension contribution degree of the dimension nodes in the hierarchical topology network.

[0180] Thus, according to the feature types in the target product data, the data attribution result set can be determined according to the data attribution recognition model and the live dimension link in a targeted manner.

[0181] ​Therefore, by fusing the JS divergence, the index dimension ascending / descending sequence of the sequence pattern, and the data attribution method of the dimension characteristic correlation, the method determines the data analysis attribution relationship by constructing the index dimension ascending / descending sequence and the dimension correlation characteristics, fusing the index dimension link constructed by the JS divergence and the dimension contribution degree, and thus accurately and quickly locates the dimensions of the influence index abnormality. The method has wide application value and reference significance in related scenes of data attribution. The method is also applicable to other data analysis and mining scenes, that is, the data attribution method fusing the JS divergence, the sequence pattern, and the characteristic correlation is within the protection scope of the present application.

[0182] In the foregoing Figure 1 - Figure 6b Based on the embodiments of the present application, Figure 7 A device structure diagram of a data processing apparatus provided by the embodiments of the present application is shown in FIG. 7. The data processing apparatus 700 includes an acquisition unit 701, a determination unit 702, and an identification unit 703.

[0183] The acquisition unit 701 is configured to acquire historical product data related to a target product.

[0184] The determination unit 702 is configured to determine a data index corresponding to the historical product data, and a data dimension characteristic corresponding to the data index in the historical product data.

[0185] The determination unit 702 is further configured to determine, according to the data index and the data dimension characteristic corresponding thereto, an index dimension ascending / descending sequence of the data index corresponding to a plurality of continuous time windows, and determine, for an i th time window of the plurality of time windows, an index dimension ascending / descending sequence corresponding to the i th time window, which is used to identify ascending / descending information of the data index and the data dimension characteristic corresponding thereto in the i th time window compared to an (i-1) th time window.

[0186] The determination unit 702 is further configured to train a data attribution identification model according to the index dimension ascending / descending sequence and an actual data attribution label of the data index determined based on the historical product data.

[0187] The acquisition unit 702 is further configured to acquire target product data of the target product, the target product data being determined according to a to-be-identified data index.

[0188] The identification unit 703 is configured to determine, by the data attribution identification model, a data attribution result corresponding to the to-be-identified data index according to the target product data.

[0189] In a possible implementation manner, the determination unit is further configured to:

[0190] The multiple time windows are divided into multiple investigation periods based on the investigation duration, and the investigation periods include N time windows;

[0191] According to the ascending and descending sequence of the index dimension corresponding to the kth time window in each of the at least two investigation periods, a longest prefix sequence of the data index in the kth time window is determined according to a minimum support threshold, wherein the data dimension feature included in the longest prefix sequence satisfies the minimum support threshold in that the same ascending and descending information appears in the at least two investigation periods, and k is an element of N;

[0192] According to the longest prefix sequence, a data attribution recognition model is trained based on the actual data attribution label.

[0193] In a possible implementation, the determining unit is further configured to:

[0194] According to the longest prefix sequence of the data index corresponding to each of the investigation periods, a dimension feature correlation degree of the data index between the N time windows is determined based on a multi-head attention model.

[0195] According to the dimension feature correlation degree and the longest prefix sequence, a data attribution recognition model is trained based on the actual data attribution label.

[0196] In a possible implementation, the determining unit is further configured to:

[0197] A feature type of the data dimension feature corresponding to the data index is determined, and the feature type includes a time sequence type of a feature numerical value changing over time and a non-time sequence type.

[0198] If it is determined that the data dimension feature includes a data dimension feature of the time sequence type, for the data dimension feature of the time sequence type, the determining unit is further configured to determine an index dimension ascending and descending sequence of the data index corresponding to the multiple time windows according to the numerical value change of the data index and the corresponding data dimension feature in the continuous multiple time windows.

[0199] In a possible implementation, the apparatus further includes a generating unit:

[0200] If it is determined that the data dimension feature includes a target dimension feature of the non-time sequence type, the generating unit is configured to generate a hierarchical topology network between the data index and the target dimension feature according to a parent-child relationship between the target dimension features in the data dimension, and a dimension node in the hierarchical topology network is used to identify the target dimension feature.

[0201] The determining unit is further configured to determine, as a candidate node, a dimension node whose JS divergence and dimension contribution degree meet preset conditions respectively, according to the JS divergence and the dimension contribution degree of the dimension node, wherein the JS divergence is used to identify the probability distribution correlation between the dimension node and a parent node, and the dimension contribution degree is used to identify the contribution degree of a child node of the dimension node to the dimension node.

[0202] The determining unit is further configured to determine, as an indicator dimension link, a data dimension characteristic having a data attribution relationship with the data indicator, according to the candidate node in the hierarchical topology network based on a parent-child relationship.

[0203] In a possible implementation, the hierarchical topology network includes a plurality of levels, dimension nodes at a same level have a same path step number to a corresponding node of the data indicator, and for a jth level and a (j+1)th level in the plurality of levels, the determining unit is further configured to:

[0204] determine a JS divergence and a dimension contribution degree of a dimension node at the jth level respectively;

[0205] determine a candidate node of the jth level according to the preset conditions;

[0206] determine a candidate node of the (j+1)th level according to the preset conditions based on a child node of the candidate node of the jth level in the (j+1)th level.

[0207] In a possible implementation, the determining unit is further configured to:

[0208] determine a first to-be-identified data dimension characteristic of a non-temporal type in the target product data;

[0209] determine a first data attribution result corresponding to the to-be-identified data indicator in the first to-be-identified data dimension characteristic according to the indicator dimension link.

[0210] In a possible implementation, the identifying unit is further configured to:

[0211] determine a second to-be-identified data dimension characteristic of a temporal type in the target product data;

[0212] determine a second data attribution result corresponding to the to-be-identified data indicator according to the second to-be-identified data dimension characteristic through the data attribution identification model;

[0213] determine a data attribution result set for the to-be-identified data indicator according to the first data attribution result and the second data attribution result.

[0214] It can be seen that for the target product that needs data attribution, the historical product data related to it is obtained, and the numerical changes of the data indicators and data dimension features in multiple time windows are determined based on the data indicators and corresponding data dimension features in the historical product data. Based on the numerical changes, an indicator dimension rising and falling sequence with time window as the granularity is obtained. Since the indicator dimension rising and falling sequence identifies the numerical rise and fall information of the data indicators and data dimension features in different time windows, it can reflect the changes in the correlation between the data indicators and data dimension features under time changes. Therefore, according to the indicator dimension rising and falling sequence and the actual data attribution labels of the data indicators determined based on the historical product data, the trained data attribution recognition model can identify the data attribution results for the data indicators to be identified in different time windows based on the target product data. Compared with fixed expert rules, the data attribution recognition model can provide more dynamic data attribution results in the time dimension, which can adapt to the needs of dynamic operation and maintenance changes. The extraction of data dimension features associated with data indicators based on the data attribution results is more flexible.

[0215] The present application also provides a computer device, which is the aforementioned computer device and may include a terminal device or a server. The aforementioned data processing device may be configured in the computer device. The computer device is described below with reference to the accompanying drawings.

[0216] If the computer device is a terminal device, see Figure 8 As shown, the embodiment of the present application provides a terminal device, taking a mobile phone as an example:

[0217] Figure 8 The block diagram shows a partial structure of a mobile phone related to the terminal device provided in the embodiment of the present application. Figure 8 The mobile phone includes components such as a radio frequency (RF) circuit 1410, a memory 1420, an input unit 1430, a display unit 1440, a sensor 1450, an audio circuit 1460, a wireless fidelity (WiFi) module 1470, a processor 1480, and a power supply 1490. It will be understood by those skilled in the art that Figure 8 The mobile phone structure shown in the figure does not constitute a limitation to the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0218] The following combination Figure 8 A detailed introduction to the various components of a mobile phone:

[0219] The RF circuit 1410 can be used for receiving and sending signals in the process of information or communication, in particular, receiving the downlink information from the base station and processing it by the processor 1480, and sending the uplink data to the base station. Generally, the RF circuit 1410 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (LNA), a duplexer, etc. In addition, the RF circuit 1410 can also communicate with the network and other devices through wireless communication. The above-mentioned wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile Communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), Email, Short Messaging Service (SMS), etc.

[0220] The memory 1420 can be used to store software programs and modules, and the processor 1480 can execute various functions of the mobile phone and data processing by running the software programs and modules stored in the memory 1420. The memory 1420 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory 1420 can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device.

[0221] The input unit 1430 can be used to receive input digital or character information, and to generate key signal input related to the user settings and function control of the mobile phone. Specifically, the input unit 1430 may include a touch panel 1431 and other input devices 1432. The touch panel 1431, also known as a touch screen, can collect user touch operations on or near it (such as operations performed by the user using a finger, stylus, or any other suitable object or accessory on or near the touch panel 1431) and drive the corresponding connection device according to a pre-set program. Optionally, the touch panel 1431 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch direction and detects the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device and converts it into touch point coordinates, which are then sent to the processor 1480. It can also receive commands sent by the processor 1480 and execute them. In addition, the touch panel 1431 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 1431, the input unit 1430 may further include other input devices 1432. Specifically, the other input devices 1432 may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control keys, power keys, etc.), a trackball, a mouse, and a joystick.

[0222] The display unit 1440 can be used to display information input by the user or information provided to the user and various menus of the mobile phone. The display unit 1440 may include a display panel 1441. Optionally, the display panel 1441 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. Further, the touch panel 1431 may cover the display panel 1441. When the touch panel 1431 detects a touch operation on or near it, it is transmitted to the processor 1480 to determine the type of touch event. Subsequently, the processor 1480 provides corresponding visual output on the display panel 1441 according to the type of touch event. Although in Figure 8 In the embodiment, the touch panel 1431 and the display panel 1441 are used as two independent components to realize the input and output functions of the mobile phone, but in some embodiments, the touch panel 1431 and the display panel 1441 can be integrated to realize the input and output functions of the mobile phone.

[0223] The phone can also include at least one sensor 1450, such as an optical sensor, a motion sensor, and other sensors. Specifically, the optical sensor can include an ambient light sensor to adjust the brightness of the display panel 1441 according to the brightness of ambient light, and a proximity sensor to turn off the display panel 1441 and / or the backlight when the phone is moved to the ear. As one of the motion sensors, the accelerometer sensor can detect the magnitude of acceleration in each direction (usually three axes), and when at rest, it can detect the magnitude and direction of gravity, which can be used for applications that identify the phone posture (such as switching between landscape and portrait screens, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometers, taps), and the like. As for other sensors that the phone can also be configured, such as a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, and the like, they will not be described here.

[0224] The audio circuit 1460, the speaker 1461, and the microphone 1462 can provide an audio interface between the user and the phone. The audio circuit 1460 can convert the received audio data into an electrical signal, transmit it to the speaker 1461, and convert it into a sound signal output by the speaker 1461; on the other hand, the microphone 1462 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1460 and converted into audio data, which is then processed by the processor 1480 and transmitted to another phone via the RF circuit 1410, or output to the memory 1420 for further processing.

[0225] WiFi is a short-range wireless transmission technology. The WiFi module 1470 can help users send and receive emails, browse web pages, and access streaming media, etc. It provides users with wireless broadband Internet access. Although Figure 8 The WiFi module 1470 is shown, but it is understood that it does not belong to the necessary components of the phone, and can be omitted as needed without changing the essence of the invention.

[0226] The processor 1480 is the control center of the phone, which connects all parts of the phone through various interfaces and lines, and performs various functions and processes data of the phone by running or executing software programs and / or modules stored in the memory 1420, and calling data stored in the memory 1420. Optionally, the processor 1480 can include one or more processing units; preferably, the processor 1480 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 1480.

[0227] The mobile phone further includes a power supply 1490 (such as a battery) for supplying power to each component. Preferably, the power supply is logically connected to the processor 1480 through a power management system, so that the power management system can realize functions such as charge management, discharge management, and power consumption management.

[0228] Although not shown, the mobile phone can further include a camera, a Bluetooth module, and the like, which will not be described here.

[0229] In this embodiment, the processor 1480 included in the terminal device further has the following functions:

[0230] obtain historical product data related to a target product;

[0231] determine a data index corresponding to the historical product data, and a data dimension feature corresponding to the data index in the historical product data;

[0232] According to the data index and the corresponding data dimension feature, determine the index dimension ascending / descending sequence corresponding to the data index in the plurality of time windows, and for the i th time window of the plurality of time windows, the index dimension ascending / descending sequence corresponding to the i th time window is used to identify the ascending / descending information of the data index and the corresponding data dimension feature in the i th time window compared with the i-1 th time window;

[0233] According to the index dimension ascending / descending sequence and the actual data attribution label of the data index determined based on the historical product data, a data attribution recognition model is trained;

[0234] obtain target product data of the target product, the target product data being determined according to a to-be-recognized data index;

[0235] According to the target product data, determine the data attribution result corresponding to the to-be-recognized data index through the data attribution recognition model.

[0236] If the computer device is a server, the embodiments of the present application further provide a server, please refer to Figure 9 shown, Figure 9 The structural diagram of the server 1500 provided in the embodiment of the present application, the server 1500 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPUs) 1522 (for example, one or more processors) and a memory 1532, and one or more storage media 1530 (for example, one or more mass storage devices) for storing application programs 1542 or data 1544. Among them, the memory 1532 and the storage medium 1530 can be temporary storage or permanent storage. The program stored in the storage medium 1530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1522 can be configured to communicate with the storage medium 1530 to execute a series of instruction operations in the storage medium 1530 on the server 1500.

[0237] The server 1500 may also include one or more power supplies 1526, one or more wired or wireless network interfaces 1550, one or more input and output interfaces 1558, and / or one or more operating systems 1541, such as Windows Server 2003. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM etc.

[0238] The steps performed by the server in the above embodiment can be based on Figure 9 The server structure shown.

[0239] In addition, an embodiment of the present application further provides a storage medium, which is used to store a computer program, and the computer program is used to execute the method provided by the above embodiment.

[0240] An embodiment of the present application also provides a computer program product including instructions, which, when executed on a computer, enables the computer to execute the method provided in the above embodiment.

[0241] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the above-mentioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the above-mentioned storage medium can be at least one of the following media: read-only memory (English: Read-only Memory, abbreviated: ROM), RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0242] It should be noted that each of the embodiments described in the specification of the present application adopts a progressive description manner, and the same or similar parts between the embodiments can be mutually referred to. Each of the embodiments focuses on the differences from other embodiments. In particular, the device and system embodiments are described more simply because they are basically similar to the method embodiments, and the relevant parts can be referred to the part of the method embodiments. The above-described device and system embodiments are merely illustrative, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiments according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0243] The above describes only one specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application. Moreover, the implementation manners provided in the above aspects of the present application can be further combined to provide more implementation manners. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A data processing method, characterized in that: The method comprises: Obtain historical product data related to the target product; Determining a data indicator corresponding to the historical product data, and a data dimension feature in the historical product data corresponding to the data indicator; If the data dimension features include time-series data dimension features, determine the indicator dimension ascending and descending sequences corresponding to the multiple time windows of the data indicators based on the numerical changes of the data indicators and the corresponding data dimension features in multiple consecutive time windows. For the i-th time window of the multiple time windows, the corresponding indicator dimension ascending and descending sequence is used to identify the ascending and descending information of the data indicators and the corresponding data dimension features in the i-th time window compared to the i-1-th time window. Mining the longest prefix sequence according to the indicator dimension ascending and descending sequence, and processing the longest prefix sequence through a multi-head attention model to determine the dimensional feature correlation of the data indicator between multiple time windows; Training a data attribution recognition model based on the correlation between the longest prefix sequence and the dimensional feature and the actual data attribution label of the data indicator determined based on the historical product data; If the data dimension features include a target dimension feature of a non-temporal type, a hierarchical topological network is generated between the data indicator and the target dimension feature, and the indicator dimension link is determined by the JS divergence and dimension contribution of the dimension nodes in the hierarchical topological network. The non-temporal type includes a feature type with continuous values ​​and unrelated to the product cycle and a feature type with discrete values; Acquiring target product data of the target product, wherein the target product data is determined based on the data indicator to be identified; According to different feature types in the target product data, a data attribution result set is determined specifically based on the data attribution identification model and the indicator dimension link.

2. The method according to claim 1, characterized in that The mining of the longest prefix sequence according to the ascending and descending sequence of the indicator dimension includes: Dividing the multiple time windows into multiple inspection periods based on the inspection duration, wherein the inspection period includes N time windows; According to the indicator dimension rising and falling sequences corresponding to the k-th time window in at least two of the inspection periods, the longest prefix sequence of the data indicator in the k-th time window is determined according to the minimum support threshold, wherein the number of times the same rising and falling information of the data dimension features included in the longest prefix sequence appears in at least two of the inspection periods satisfies the minimum support threshold, k∈N.

3. The method according to claim 2, characterized in that Processing the longest prefix sequence through a multi-head attention model to determine the dimensional feature correlation of the data indicator across multiple time windows includes: The dimensional feature correlation of the data indicators between the N time windows is determined based on the multi-head attention model through the longest prefix sequences corresponding to the data indicators in the inspection period.

4. The method according to claim 1, wherein Generating a hierarchical topological network between the data indicator and the target dimensional feature, and determining the indicator dimension link through the JS divergence and dimension contribution of the dimensional nodes in the hierarchical topological network, includes: Generate a hierarchical topological network between the data indicators and the target dimensional features according to the parent-child relationship between the target dimensional features in the data dimension, wherein the dimension nodes in the hierarchical topological network are used to identify the target dimensional features; According to the Jason Shannon JS divergence and dimension contribution of the dimension node, the dimension node whose JS divergence and dimension contribution respectively meet the preset conditions is determined as a candidate node; wherein the JS divergence is used to identify the probability distribution correlation between the dimension node and the parent node, and the dimension contribution is used to identify the contribution degree of the child node of the dimension node to the dimension node; According to the candidate nodes, based on the parent-child relationship in the hierarchical topology network, an indicator dimension link formed with the data indicator is determined, and the indicator dimension link is used to identify a data dimension feature having a data attribution relationship with the data indicator.

5. The method according to claim 4, characterized in that The hierarchical topology network includes multiple levels, and the number of path steps from the dimension nodes at the same level to the node corresponding to the data indicator is the same. For the j-th level and the j+1-th level of the multiple levels, the dimensional nodes whose JS divergence and dimensional contribution respectively meet preset conditions are determined as candidate nodes according to the Jason Shannon JS divergence and dimensional contribution of the dimension nodes, including: Determine the JS divergence and dimension contribution corresponding to the dimension node at the j-th level respectively; Determine the candidate nodes of the j-th level according to the preset conditions; Based on the child nodes of the candidate node at the jth level at the j+1th level, the candidate node at the j+1th level is determined according to the preset condition.

6. The method according to claim 4, characterized in that After acquiring the target product data of the target product, the method further includes: Determining a first to-be-identified data dimension feature of a non-time-series type in the target product data; According to the indicator dimension link, a first data attribution result corresponding to the indicator of the data to be identified in the first data dimension feature to be identified is determined.

7. The method according to claim 6, characterized in that The method further comprises: Determining a second to-be-identified data dimension feature of a time series type in the target product data; Determining, by the data attribution recognition model, a second data attribution result corresponding to the to-be-identified data indicator according to the second to-be-identified data dimensional feature; The method further comprises: A data attribution result set for the data indicator to be identified is determined based on the first data attribution result and the second data attribution result.

8. A data processing device, characterized in that: The device includes an acquisition unit, a determination unit, a generation unit and an identification unit: The acquisition unit is used to acquire historical product data related to the target product; The determining unit is configured to determine a data indicator corresponding to the historical product data, and a data dimension feature in the historical product data corresponding to the data indicator; The determining unit is further configured to, if the data dimension features include data dimension features of a time series type, determine, based on numerical changes of the data indicators and the corresponding data dimension features in a plurality of consecutive time windows, indicator dimension ascending and descending sequences of the data indicators corresponding to the plurality of time windows respectively; for an i-th time window of the plurality of time windows, the corresponding indicator dimension ascending and descending sequence is used to identify ascending and descending information of the data indicators and the corresponding data dimension features in the i-th time window compared to the i-1-th time window; mine a longest prefix sequence based on the indicator dimension ascending and descending sequence; and process the longest prefix sequence through a multi-head attention model to determine the dimensional feature correlation of the data indicators between the plurality of time windows; The determining unit is further configured to train a data attribution recognition model based on the correlation between the longest prefix sequence and the dimensional feature and the actual data attribution label of the data indicator determined based on the historical product data; The generating unit is configured to generate a hierarchical topological network between the data indicator and the target dimensional feature if the data dimensional features include a target dimensional feature of a non-temporal type, and determine an indicator dimensional link through the JS divergence and dimensional contribution of the dimensional nodes in the hierarchical topological network, wherein the non-temporal type includes a feature type having continuous values ​​and being unrelated to the product cycle and a feature type having discrete values; The acquisition unit is further configured to acquire target product data of the target product, wherein the target product data is determined based on the data indicator to be identified; The identification unit is used to determine a data attribution result set based on the data attribution identification model and the indicator dimension link according to different feature types in the target product data.

9. The device according to claim 8, characterized in that The determining unit is further configured to: Dividing the multiple time windows into multiple inspection periods based on the inspection duration, wherein the inspection period includes N time windows; According to the indicator dimension rising and falling sequences corresponding to the k-th time window in at least two of the inspection periods, the longest prefix sequence of the data indicator in the k-th time window is determined according to the minimum support threshold, wherein the number of times the same rising and falling information of the data dimension features included in the longest prefix sequence appears in at least two of the inspection periods satisfies the minimum support threshold, k∈N.

10. The device according to claim 9, characterized in that The determining unit is further configured to: The dimensional feature correlation of the data indicators between the N time windows is determined based on the multi-head attention model through the longest prefix sequences corresponding to the data indicators in the inspection period.

11. The device according to claim 8, characterized in that The determining unit is further configured to: Generate a hierarchical topological network between the data indicators and the target dimensional features according to the parent-child relationship between the target dimensional features in the data dimension, wherein the dimension nodes in the hierarchical topological network are used to identify the target dimensional features; According to the Jason Shannon JS divergence and dimension contribution of the dimension node, the dimension node whose JS divergence and dimension contribution respectively meet the preset conditions is determined as a candidate node; wherein the JS divergence is used to identify the probability distribution correlation between the dimension node and the parent node, and the dimension contribution is used to identify the contribution degree of the child node of the dimension node to the dimension node; According to the candidate nodes, based on the parent-child relationship in the hierarchical topology network, an indicator dimension link formed with the data indicator is determined, and the indicator dimension link is used to identify a data dimension feature having a data attribution relationship with the data indicator.

12. The device according to claim 11, characterized in that The determining unit is further configured to: The hierarchical topology network includes multiple levels, and the number of path steps from the dimension nodes at the same level to the nodes corresponding to the data indicators is the same, for the j-th level and the j+1-th level in the multiple levels; Determine the JS divergence and dimension contribution corresponding to the dimension node at the j-th level respectively; Determine the candidate nodes of the j-th level according to the preset conditions; Based on the child nodes of the candidate node at the jth level at the j+1th level, the candidate node at the j+1th level is determined according to the preset condition.

13. The device according to claim 11, characterized in that The determining unit is further configured to: Determining a first to-be-identified data dimension feature of a non-time-series type in the target product data; According to the indicator dimension link, a first data attribution result corresponding to the indicator of the data to be identified in the first data dimension feature to be identified is determined.

14. The device according to claim 13, characterized in that The identification unit is further configured to: Determining a second to-be-identified data dimension feature of a time series type in the target product data; Determining, by the data attribution recognition model, a second data attribution result corresponding to the to-be-identified data indicator according to the second to-be-identified data dimensional feature; A data attribution result set for the data indicator to be identified is determined based on the first data attribution result and the second data attribution result.

15. A computer device, characterized in that: The computer device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the data processing method according to any one of claims 1 to 7 according to instructions in the program code.

16. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the data processing method according to any one of claims 1 to 7.

17. A computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the data processing method according to any one of claims 1 to 7.

Images