CN115905698B - User portrait analysis method and system based on artificial intelligence - Google Patents
User portrait analysis method and system based on artificial intelligence Download PDFInfo
- Publication number
- CN115905698B CN115905698B CN202211508615.6A CN202211508615A CN115905698B CN 115905698 B CN115905698 B CN 115905698B CN 202211508615 A CN202211508615 A CN 202211508615A CN 115905698 B CN115905698 B CN 115905698B
- Authority
- CN
- China
- Prior art keywords
- network
- user preference
- debugging
- basic
- examples
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 184
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 36
- 230000000694 effects Effects 0.000 claims abstract description 114
- 230000006872 improvement Effects 0.000 claims abstract description 28
- 238000000611 regression analysis Methods 0.000 claims description 143
- 238000000034 method Methods 0.000 claims description 40
- 230000008859 change Effects 0.000 claims description 35
- 238000011156 evaluation Methods 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 abstract description 16
- 238000005065 mining Methods 0.000 abstract description 11
- 238000012216 screening Methods 0.000 abstract description 3
- 239000013598 vector Substances 0.000 description 16
- 230000008569 process Effects 0.000 description 9
- 230000000007 visual effect Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 230000028161 membrane depolarization Effects 0.000 description 6
- 230000003993 interaction Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 241000282414 Homo sapiens Species 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
According to the user portrayal analysis method and system based on artificial intelligence, when a portrayal analysis request is received, user activity big data to be analyzed is obtained based on the portrayal analysis request; and then inputting the big data of the user activities to be analyzed into an expert system analysis network for finishing improvement, and obtaining the analysis result of the user preference items output by the expert system analysis network for finishing improvement. By determining the big data of the user activity to be analyzed, which is matched with the portrait analysis request, the preliminary screening of the big data of the user activity can be realized, the data processing pressure of the improved expert system analysis network is reduced, in addition, the debugging timeliness of the improved expert system analysis network is higher, so that the analysis result of the user preference can be rapidly output, and further, the improved expert system analysis network can realize the fine mining of the user preference through improvement, so that the analysis result of the user preference can be accurately output.
Description
Technical Field
The invention relates to the technical field of user portrayal, in particular to a user portrayal analysis method and system based on artificial intelligence.
Background
Artificial intelligence is a principle of researching human intelligence activities, constructing an artificial system with a certain intelligence, researching how to make a computer complete work which needs human intelligence in the past, namely, researching basic theory, method and technology how to apply software and hardware of the computer to simulate certain intelligent behaviors of human beings. At present, based on the big data network era, no matter product planning, product operation or data pushing, the maximum visible value and invisible value of a user are needed to be obtained, and the necessary value and the value added value are needed to be obtained. The related art has applied artificial intelligence to image analysis, but the defects of high cost and low precision still exist in practical application.
Disclosure of Invention
In order to improve the technical problems in the related art, the invention provides a user portrait analysis method and a system based on artificial intelligence.
In a first aspect, an embodiment of the present invention provides an artificial intelligence-based user portrait analysis method, which is applied to an artificial intelligence-based user portrait analysis system, and the method at least includes:
When a portrait analysis request is received, acquiring big data of user activities to be analyzed based on the portrait analysis request;
and inputting the big data of the user activities to be analyzed into an expert system analysis network for finishing improvement, and obtaining a user preference analysis result output by the expert system analysis network for finishing improvement.
Optionally, the expert system analysis network for performing the improvement is improved by:
obtaining an expert system analysis network for completing debugging and a basic network debugging basis for debugging the expert system analysis network; wherein the network commissioning examples in the underlying network commissioning basis are internet user activity big data comprising a first user preference and a second user preference, and the associated activity information set of the second user preference is in the associated activity information set of the first user preference, the underlying a priori knowledge of at least part of the network commissioning examples in the underlying network commissioning basis does not enable a comprehensive annotation of the first user preference and/or the second user preference;
performing regression analysis on at least part of network debugging examples in the basic network debugging basis based on the expert system analysis network to obtain regression analysis data of the part of network debugging examples aiming at the first user preference or the second user preference;
Combining regression analysis data of the partial network debugging examples and basic priori knowledge of the corresponding network debugging examples in the basic network debugging basis to determine knowledge endowed quality information; wherein the knowledge-endowed quality information reflects whether regression analysis data of the partial network commissioning example for the first user preference or the second user preference is of a target prediction class;
and changing the basic network debugging basis based on the knowledge endowed quality information to obtain a network debugging basis for completing the change so as to improve the expert system analysis network.
Optionally, the first user preference and/or the second user preference in the basic prior knowledge are annotated by a setting manner; the expert system analysis network includes a first user preference identification network and a second user preference identification network; the method further comprises the steps of:
obtaining an unbugged underlying user preference identification network;
taking the priori knowledge of the first user preference in the basic priori knowledge as a first debugging instruction, and debugging the basic user preference identification network based on the basic network debugging basis until the first debugging requirement including maintaining network generalization indexes is met, so as to obtain the first user preference identification network;
And debugging the basic user preference identification network based on the basic network debugging basis on the basis of the priori knowledge of the first user preference and the priori knowledge of the second user preference until the second debugging requirement including maintaining network generalization indexes is met, so as to obtain the second user preference identification network.
Optionally, the step of debugging the basic user preference identification network based on the prior knowledge of the first user preference and the prior knowledge of the second user preference until meeting a second debugging requirement including maintaining a network generalization index, to obtain the second user preference identification network includes:
based on the prior knowledge of the first user preference, obtaining an associated activity information set of the first user preference covered by a corresponding network debugging example from the network debugging examples in the basic network debugging basis to obtain a first network debugging basis;
and taking the priori knowledge of the second user preference in the basic priori knowledge as a second debugging instruction, and debugging the basic user preference identification network based on the first network debugging basis until the second debugging requirement comprising maintaining a network generalization index is met, so as to obtain the second user preference identification network.
Optionally, the determining, by combining regression analysis data of the partial network debug instance and basic prior knowledge of a corresponding network debug instance in the basic network debug basis, knowledge-endowing quality information includes:
performing joint analysis on regression analysis data of the partial network debugging examples and target priori knowledge of the corresponding network debugging examples in the basic network debugging basis to obtain debugging reference comparison information; the debugging reference comparison information reflects whether regression analysis data of the partial network debugging examples are suitable for target priori knowledge of corresponding network debugging examples in the basic network debugging basis;
according to the regression analysis data of the partial network debugging examples, not adapting to the target priori knowledge of the corresponding network debugging examples in the basic network debugging basis, outputting the regression analysis data of the partial network debugging examples and the target priori knowledge of the corresponding network debugging examples to obtain derivative evaluation information; the derived evaluation information is used for reflecting whether regression analysis data of the corresponding network debugging examples in the partial network debugging examples is the target prediction category.
Optionally, the step of modifying the basic network debugging basis based on the knowledge-endowed quality information to obtain a modified network debugging basis includes:
According to the knowledge endowing quality information, reflecting that regression analysis data of the partial network debugging examples are not of the target prediction category, and changing the regression analysis data of the partial network debugging examples into priori knowledge of the corresponding network debugging examples;
and according to regression analysis data of the partial network debugging examples, the target prediction category is obtained, and the prior knowledge of the corresponding network debugging examples is unchanged.
Optionally, the step of debugging the basic user preference identification network based on the basic network debugging basis until the first debugging requirement including maintaining a network generalization index is met, and obtaining the first user preference identification network includes: identifying the basic network debugging basis based on the basic user preference identification network to obtain first user preference identification information; according to the first comparison data between the basic priori knowledge and the first user preference identification information does not accord with a first requirement, optimizing network model variables of the basic user preference identification network to obtain the first user preference identification network;
the step of debugging the basic user preference identification network based on the basic network debugging basis until the basic user preference identification network meets a second debugging requirement including maintaining network generalization indexes, and obtaining the second user preference identification network comprises the following steps: identifying the first network debugging basis based on the basic user preference identification network to obtain second user preference identification information; and optimizing network model variables of the basic user preference identification network according to the fact that second comparison data between the basic priori knowledge and the second user preference identification information does not meet a second requirement, so as to obtain the second user preference identification network.
Optionally, the method further comprises:
performing regression analysis on at least part of network debugging examples in the network debugging basis completing the change based on the expert system analysis network to obtain regression analysis data of at least part of network debugging examples in the network debugging basis completing the change aiming at the first user preference or the second user preference;
and finishing the improvement of the expert system analysis network on the basis that regression analysis data of the network debugging examples with the set duty ratio in the changed network debugging basis is not in the target prediction category.
In a second aspect, the invention also provides a user portrait analysis system based on artificial intelligence, which comprises a processor and a memory; the processor is in communication with the memory, and the processor is configured to read and execute a computer program from the memory to implement the method described above.
In a third aspect, the present invention also provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the method described above.
With the embodiments of the present invention, at least the following advantageous effects can be achieved.
(1) When a portrait analysis request is received, acquiring large data of user activities to be analyzed based on the portrait analysis request; and then inputting the big data of the user activities to be analyzed into an expert system analysis network for finishing improvement, and obtaining the analysis result of the user preference items output by the expert system analysis network for finishing improvement. By determining the big data of the user activity to be analyzed, which is matched with the portrait analysis request, the preliminary screening of the big data of the user activity can be realized, the data processing pressure of the improved expert system analysis network is reduced, in addition, the debugging timeliness of the improved expert system analysis network is higher, so that the analysis result of the user preference can be rapidly output, and further, the improved expert system analysis network can realize the fine mining of the user preference through improvement, so that the analysis result of the user preference can be accurately output.
(2) The first user preference and/or the second user preference according to the partial network debugging example are debugged through the partial annotation basic network to obtain basic priori knowledge, and the basic network debugging basis is used as a debugging expert system analysis network in view of the fact that the resource consumption of the first user preference and/or the second user preference is reduced, so that the annotation time consumption can be reduced, the non-debugged basic user preference identification network is debugged through the partial annotation basic network debugging basis to obtain the expert system analysis network, and the debugging timeliness of the expert system analysis network can be improved based on the saving of the annotation time consumption.
(3) Regression analysis can be performed on at least part of network debugging examples in the basic network debugging basis based on the expert system analysis network to obtain regression analysis data of the part of network debugging examples about the first user preference or the second user preference, so that knowledge endowing quality information can be determined based on the regression analysis data of the part of network debugging examples and basic priori knowledge of corresponding network debugging examples in the basic network debugging basis; the knowledge endowing quality information reflects whether the regression analysis data of the first user preference or the second user preference is of a target prediction type or not according to part of the network debugging examples, so that whether the regression analysis data of the expert system analysis network is of the target prediction type or not can be timely and accurately determined, and the judging timeliness of error correction of the regression analysis data is improved.
(4) The quality information can be endowed on the basis of knowledge to change the basic network debugging basis, and the changed network debugging basis is obtained to improve the expert system analysis network, so that the accuracy and timeliness of user preference annotation can be further improved, the number of user preference annotation is increased, and the accuracy and anti-interference performance of the expert system analysis network are further improved on the basis of multiple rounds of improvement. In addition, in view of the fact that the associated activity information set of the second user preference is in the associated activity information set of the first user preference, the expert system analysis network can pertinently mine and identify the second user preference based on the associated activity information set of the first user preference on the basis of identifying the associated activity information set of the first user preference, the defect that the second user preference cannot be identified due to the fact that the information set of the second user preference is fewer is avoided, and accuracy and credibility of user preference identification are guaranteed.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
FIG. 1 is a flow chart of a user portrait analysis method based on artificial intelligence according to an embodiment of the present invention.
FIG. 2 is a schematic diagram of a communication architecture of an application environment of an artificial intelligence-based user portrait analysis method according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the invention. Rather, they are merely examples of apparatus and methods consistent with aspects of the invention as detailed in the accompanying claims.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.
The method embodiments provided by the embodiments of the present invention may be implemented in an artificial intelligence based user profile analysis system, a computer device or similar computing device. Taking the example of operating on an artificial intelligence based user profile analysis system, the artificial intelligence based user profile analysis system 10 may include one or more processors 102 (the processor 102 may include, but is not limited to, a microprocessor MCU, a programmable logic device FPGA, etc. processing means) and a memory 104 for storing data, and optionally the artificial intelligence based user profile analysis system may also include a transmission means 106 for communication functions. It will be appreciated by those of ordinary skill in the art that the above-described architecture is merely illustrative and is not intended to limit the architecture of the artificial intelligence-based user profile analysis system described above. For example, artificial intelligence based user profile analysis system 10 may also include more or fewer components than those shown above, or have a different configuration than those shown above.
The memory 104 may be used to store a computer program, for example, a software program of application software and a module, such as a computer program corresponding to an artificial intelligence-based user portrait analysis method in an embodiment of the present invention, and the processor 102 executes the computer program stored in the memory 104 to perform various functional applications and data processing, that is, implement the method described above. Memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, memory 104 may further include memory remotely located with respect to processor 102, which may be connected to artificial intelligence based user profile analysis system 10 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The transmission means 106 is arranged to receive or transmit data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider of the artificial intelligence based user profile analysis system 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, simply referred to as NIC) that can connect to other network devices through a base station to communicate with the internet. In one example, the transmission device 106 may be a Radio Frequency (RF) module, which is used to communicate with the internet wirelessly.
Based on this, referring to fig. 1, fig. 1 is a schematic flow chart of a user portrait analysis method based on artificial intelligence according to an embodiment of the present invention, where the method is applied to a user portrait analysis system based on artificial intelligence, and further may include the technical solutions described in step 1 and step 2.
And step 1, when a portrait analysis request is received, acquiring big data of user activities to be analyzed based on the portrait analysis request.
The portrait analysis request can be sent by a third party service platform system and is used for requesting the portrait/preference analysis of the user based on the artificial intelligence user portrait analysis system, and the portrait analysis system based on the artificial intelligence is known and authorized by a service user end related to the big data of the user activity to be analyzed before acquiring the big data of the user activity to be analyzed, so that the process of acquiring the big data of the user activity to be analyzed based on the artificial intelligence user portrait analysis system is standard and legal.
And 2, inputting the big data of the user activities to be analyzed into an expert system analysis network for finishing improvement, and obtaining a user preference analysis result output by the expert system analysis network for finishing improvement.
The expert system analysis network can be a neural network model built by an expert system branch based on artificial intelligence, and can simulate analysis decision thinking of experts in the industry field, so that mining analysis is carried out on big data of user activities to be analyzed, and analysis results of user preference items are obtained.
The method is applied to the steps 1 and 2, the preliminary screening of the user activity big data can be realized by determining the user activity big data to be analyzed, which is matched with the portrait analysis request, the data processing pressure of finishing the improved expert system analysis network is reduced, in addition, the debugging timeliness of finishing the improved expert system analysis network is higher, so that the user preference analysis result can be rapidly output, and further, the improved expert system analysis network can realize the refined user preference mining through improvement, so that the user preference analysis result can be accurately output.
In the embodiment of the present invention, the quality and quality of the improvement of the expert system analysis network are key to guaranteeing the accuracy and reliability of the implementation of the overall scheme, so the improvement of the expert system analysis network is exemplarily described later, for example, the improvement method of the expert system analysis network may include what is described in the NODE101-NODE 104.
NODE101, obtains an expert system analysis network for completing debugging and serves as a basic network debugging basis for debugging the expert system analysis network.
The expert system analysis network may be understood as an AI neural network built in advance, such as a deep learning network, for user preference mining. Network model variables (configuration parameters) of an unretired expert system analysis network (such as a basic/initial/general user preference identification network) can be initialized in advance, the unretired expert system analysis network is debugged based on basic network debugging basis with basic priori knowledge, and therefore the network model variables are changed, and the expert system analysis network which completes the debugging is obtained. The basic network debugging basis can be understood as an original network debugging basis which is not changed/updated, the basic network debugging basis can comprise network debugging examples and basic priori knowledge of the network debugging examples, and the categories of the network debugging examples can be text categories. The basic priori knowledge is based on the priori knowledge corresponding to the network debugging examples in the network debugging basis, and by annotating the network debugging examples, whether the prior knowledge (annotation information or annotation result) such as user preference, category of the user preference and distribution characteristics of the user preference exists in the network debugging examples such as text or graphics and texts can be determined. The network commissioning examples are typically annotated in a visual window, which may be understood as a window used in the underlying a priori knowledge to reflect the distribution and size of the first user preference or the second user preference.
For some examples, changes may be made to the underlying network debug evidence, resulting in an underlying network debug evidence (sample set) that completes the change. Such as: the method comprises the steps of annotating user demand preference items in a set service session in a plurality of groups of network debugging examples (samples), taking the plurality of groups of network debugging examples annotated with the user demand preference items as basic network debugging basis, annotating GUI activity preference items in target service links in the plurality of groups of network debugging examples, taking the plurality of groups of network debugging examples annotated with the user demand preference items and the GUI activity preference items in the target service links as basic network debugging basis for completing the change, and the like. Basic network commissioning relies on a set of multiple network commissioning examples that can be used to commission a basic user preference identification network for carrying a priori knowledge. The basic network debugging basis can comprise a network debugging set and a network evaluation set, wherein the network debugging set is used for debugging the basic user preference identification network to obtain an expert system analysis network, and the network evaluation set is used for evaluating mining analysis, precision and the like of the expert system analysis network.
For some examples, the network commissioning examples in the underlying network commissioning basis may be internet user activity big data containing a first user preference and a second user preference, and the categories of the first user preference and the second user preference may be different. Such as: the first user preference is a business process guiding preference, and the second user preference is a visual interaction preference; or the first user preference is a product preference, the second user preference is an information security preference, etc. The internet user activity big data can be understood as recorded information of a user in a series of business interaction processes, and can be used as a network debugging example in a basic network debugging basis. Internet user activity big data as an example of network debugging contains user preferences that can be annotated, which can include different types of portrait preferences or points of interest.
For some examples, the associated activity information set of the second user preference may be within the associated activity information set of the first user preference, which may be understood as the content set of the internet user activity big data containing the first user preference, and the associated activity information set of the second user preference may be understood as the content set of the internet user activity big data containing the second user preference. The first user preference may be understood as a user preference with a larger coverage area and a larger number of user preferences in the internet user activity big data, and the second user preference may be understood as a user preference with a smaller coverage area and a smaller number of user preferences in the internet user activity big data, and the first user preference and the second user preference may have overlapping content sets.
For some examples, the underlying a priori knowledge of at least some of the network commissioning examples in the underlying network commissioning basis may not enable a comprehensive annotation of the first user preference and/or the second user preference. Wherein for part of the internet user activity big data, annotating all the first user preference and the second user preference, and for the rest of the internet user activity big data, not annotating the first user preference and the second user preference; alternatively, for a portion of the internet user activity profile, annotating the first user preference and the second user preference of the portion, and for the remaining internet user activity profile, not annotating the first user preference and the second user preference; alternatively, for all internet user activity big data, the first user preference and the second user preference of the annotation section, etc. Such as: the basic network debugging basis comprises 100 groups of internet user activity big data, wherein each internet user activity big data comprises at least two first user preference items and at least two second user preference items, and one first user preference item and one second user preference item in each internet user activity big data can be annotated as basic priori knowledge. The basic priori knowledge may be understood as the priori knowledge of one network debug instance in the basic network debug basis, or may be understood as the priori knowledge of all network debug instances in the basic network debug basis, e.g. each network debug instance may correspond to one basic priori knowledge.
For some examples, at least a portion of the first user preferences and at least a portion of the second user preferences in at least a portion of the internet user activity big data may be annotated based on existing annotation tag processing algorithms to obtain basic prior knowledge, and further a basic network debug basis comprising multiple sets of the internet user activity big data and the basic prior knowledge may be obtained.
And performing regression analysis on at least part of network debugging examples in the basic network debugging basis based on the expert system analysis network to obtain regression analysis data of the part of network debugging examples aiming at the first user preference or the second user preference.
In the embodiment of the invention, on the basis of obtaining the expert system analysis network which completes the debugging and the basic network debugging basis used as the debugging expert system analysis network, at least part of network debugging examples can be obtained from the basic network debugging basis at will. At least part of the network commissioning examples may be understood as a set number of internet user activity big data, may include a part of internet user activity big data in the base network commissioning basis, or may include all internet user activity big data in the base network commissioning basis.
For some examples, regression analysis (predictive processing) may be performed on some network debug examples based on the expert system analysis network completing the debug, resulting in regression analysis data (predicted results). The regression analysis data may include information such as whether the first user preference or the second user preference is identified, a category of the first user preference or the second user preference, and a distribution and a size of a regression analysis window of the first user preference or the second user preference based on the identification of the first user preference or the second user preference.
And the NODE103 is used for determining knowledge endowing quality information by combining regression analysis data of the part of network debugging examples and basic priori knowledge of the corresponding network debugging examples in the basic network debugging basis.
In the embodiment of the present invention, a mapping list/mapping relationship/correspondence between each network debug instance (for example, each internet user activity big data) and basic priori knowledge corresponding to the network debug instance may be configured in advance, for example: the first internet user activity big data corresponds to the first basic priori knowledge and the second internet user activity big data corresponds to the second basic priori knowledge. Based on the regression analysis data of the partial network debugging examples, the basic priori knowledge of the partial network debugging examples (such as the corresponding network debugging examples in the basic network debugging basis) can be obtained from the basic network debugging basis according to the partial network debugging examples which are screened currently and the mapping list between the network debugging examples and the basic priori knowledge. On the basis of obtaining the regression analysis data of the partial network debug example and the basic prior knowledge of the corresponding network debug example, a distinction between the regression analysis data of the partial network debug example and the basic prior knowledge of the corresponding network debug example may be determined. Knowledge-endowed quality information may then be determined from the distinction, the knowledge-endowed quality information may reflect whether the regression analysis data of the portion of the network debug instance with respect to the first user preference or the second user preference is of the target prediction category.
Such as: the regression analysis data of the partial network commissioning example contains 20 first user preferences, 30 second user preferences, the underlying a priori knowledge of the corresponding network commissioning example contains 26 first user preferences, 40 second user preferences, etc. Further, it may be determined whether a first user preference in the regression analysis data is present the same as a first user preference in the underlying prior knowledge and a second user preference in the regression analysis data is present the same as a second user preference in the underlying prior knowledge. Such as: manual comparison is used to determine that 20 first user preferences are the same and 28 second user preferences are the same, in other words, 6 additional first user preferences are added to the regression analysis data and 10 additional second user preferences are added.
In view of the possible bias in regression analysis data obtained through the expert system analysis network that completes the debugging, the regression analysis data is not completely correct, and for further improving the expert system analysis network and improving the accuracy of the regression analysis data later, the rationality of the additionally added first user preference and second user preference may be further determined. Such as: a first depolarization description vector of the first plurality of user preferences and a second depolarization description vector of the second plurality of user preferences annotated in the underlying prior knowledge (which may be understood as averaged feature information) may be determined, wherein the description vector of the first user preference and the description vector of the second user preference may be reflected in a list of description vectors (feature matrix). A description vector for each first user preference additionally added to the regression analysis data and a description vector for each second user preference are determined. Determining regression analysis data of the additionally added first user preference item as a positive sample class (such as a true value class) on the basis that a difference value between the description vector of the additionally added first user preference item and the first depolarization description vector belongs to a first decision value interval; determining regression analysis data of the additionally added first user preference item as a target prediction category (such as a negative sample category) on the basis that a difference value between the description vector of the additionally added first user preference item and the first depolarization description vector does not belong to the first decision value interval; determining regression analysis data of the additionally added second user preference item as a positive sample class (such as a true value class) on the basis that a difference value between the description vector of the additionally added second user preference item and the second depolarization description vector belongs to a second decision value interval; on the basis that the difference between the description vector of the additional second user preference and the second depolarization description vector does not belong to the first decision value interval, determining regression analysis data of the additional second user preference as a target prediction category (such as a negative sample category).
For some embodiments, the network debug examples in the basic network debug basis may be split to obtain multiple groups of local basic network debug bases, and regression analysis may be performed on the network debug examples in the first group of local basic network debug bases based on the expert system analysis network to obtain regression analysis data corresponding to the first group of local basic network debug bases. And then, the quality information can be given by determining knowledge between the first group of local basic network debugging basis corresponding regression analysis data and the first group of local basic network debugging basis corresponding basic priori knowledge while carrying out regression analysis on the second group of local basic network debugging basis based on the expert system analysis network. In this way, the determination efficiency of the knowledge-added quality information and the like can be improved, and the knowledge-added quality information can be understood as evaluation information of the annotation processing/labeling processing for evaluating the merits of the annotation processing/labeling processing.
NODE104, based on the knowledge endows quality information to change the basic network debugging basis, and obtains the network debugging basis for completing the change so as to improve the expert system analysis network.
Further, after determining whether the regression analysis data of the partial network debugging examples about the first user preference or the second user preference is the target prediction category, the regression analysis data which is not the target prediction category may be added to the basic priori knowledge to obtain the basic priori knowledge of the completed change, and then the basic priori knowledge of the completed change and the network debugging examples in the basic network debugging bases may be determined as the network debugging bases of the completed change. Based on the determination of the network debug evidence upon which the change is completed, the expert system analysis network may be further improved based on the network debug evidence upon which the change is completed.
Such as: and carrying out regression analysis on the network debugging examples in the changed network debugging basis based on the expert system analysis network aiming at the first user preference and the second user preference to obtain current regression analysis data. According to the current regression analysis data and the basic priori knowledge of the completion of the change, determining the current network cost/loss function of the expert system analysis network, and according to the current network cost, changing the network model variable of the expert system analysis network, thereby obtaining an improved expert system analysis network.
In the embodiment of the invention, the first user preference and/or the second user preference according to the partial network debugging example are debugged through the partial annotation basic network to obtain the basic priori knowledge, and the basic network debugging basis is used as a debugging expert system analysis network, so that the annotation time consumption can be reduced by reducing the resource consumption of the first user preference and/or the second user preference, and the non-debugged basic user preference identification network is debugged through the partial annotation basic network debugging basis to obtain the expert system analysis network, so that the debugging timeliness of the expert system analysis network can be improved based on the saving of the annotation time consumption.
Further, regression analysis can be performed on at least part of network debugging examples in the basic network debugging basis based on the expert system analysis network to obtain regression analysis data of the part of network debugging examples about the first user preference or the second user preference, so that knowledge endowing quality information can be determined based on the regression analysis data of the part of network debugging examples and basic priori knowledge of corresponding network debugging examples in the basic network debugging basis; the knowledge endowing quality information reflects whether the regression analysis data of the first user preference or the second user preference is of a target prediction type or not according to part of the network debugging examples, so that whether the regression analysis data of the expert system analysis network is of the target prediction type or not can be timely and accurately determined, and the judging timeliness of error correction of the regression analysis data is improved.
In addition, the quality information can be endowed based on knowledge to change the basic network debugging basis, and the changed network debugging basis is obtained to improve the expert system analysis network, so that the accuracy and timeliness of user preference annotation can be further improved, the number of user preference annotation is increased, and the accuracy and anti-interference performance of the expert system analysis network are further improved based on multiple rounds of improvement. Meanwhile, as the associated activity information set of the second user preference is in the associated activity information set of the first user preference, the expert system analysis network can pertinently mine and identify the second user preference based on the associated activity information set of the first user preference on the basis of identifying the associated activity information set of the first user preference, so that the defect that the second user preference cannot be identified due to the fact that the information set of the second user preference is less is avoided, and the mining precision and reliability of the user preference are improved.
For some examples, the first user preference and/or the second user preference in the underlying prior knowledge are annotated by way of a setting; the expert system analysis network includes a first user preference identification network and a second user preference identification network; the artificial intelligence based user profile analysis method may further include the following NODE111-NODE 113.
NODE111, obtains the unbuned base user preferences identifying network.
In the embodiment of the present invention, the basic user preference identification network may be understood as an untrimmed expert system analysis network, that is, a general or initial user preference identification network, which may be a convolutional neural network or a deep learning network, and the expert system analysis network may include a first user preference identification network and a second user preference identification network, where the first user preference identification network may be used to identify mining first user preferences, and the second user preference identification network may be used to identify mining second user preferences. The method for generating the basic user preference identification network can be determined according to the prior art, and the user preference identification network with the same identification category as the identification category of the internet user activity big data can be obtained from the internet platform system and used as the basic user preference identification network, so that the debugging timeliness of the expert system analysis network can be improved conveniently.
NODE112 uses a priori knowledge of a first user preference in the basic priori knowledge as a first debugging instruction, debugs the basic user preference identification network based on the basic network debugging basis until meeting a first debugging requirement including maintaining a network generalization index, and obtains the first user preference identification network.
In an embodiment of the present invention, the first user preference and/or the second user preference in the basic prior knowledge may be annotated by a setting (e.g., manually noted). The first commissioning indication may be understood as a priori knowledge that is used to commission the first user preference identification network, and the first commissioning indication may comprise a priori knowledge of the first user preference in the underlying a priori knowledge. On the basis of obtaining the basic user preference identification network, regression analysis can be performed on the network debugging examples in the basic network debugging basis based on the basic user preference identification network to obtain current regression analysis data, the current network cost of the basic user preference identification network is determined based on the current regression analysis data and the first debugging instruction, and the network model variables of the basic user preference identification network are optimized based on the current network cost, so that the first user preference identification network is obtained.
In order to reduce the probability of network generalization performance degradation/overfitting of the first user preference identification network when the basic user preference identification network is debugged, a first debugging requirement can be configured in advance, and the first debugging requirement can be understood as a debugging requirement for the first user preference identification network when the first user preference identification network is debugged. Such as: the first debug requirement is that the current network cost is less than the first stability determination value. On the basis of determining that the current network cost of the first user preference identification network is smaller than the first stability judgment value, the debugging base user preference identification network can be ended, so that the current regression analysis data and the first debugging instruction are not completely consistent, the defect that the network generalization performance of the first user preference identification network is reduced is avoided, and further, the debugging requirement can be understood as a convergence condition.
NODE113, based on the prior knowledge of the first user preference and the prior knowledge of the second user preference, debugs the basic user preference identification network based on the basic network debugs until meeting a second debugs requirement including maintaining network generalization index, and obtains the second user preference identification network.
In the embodiment of the invention, on the basis of obtaining the basic user preference identification network, as the associated activity information set of the second user preference is in the associated activity information set of the first user preference, for the second user preference, the associated activity information set of the first user preference can be obtained from the network debugging example based on the prior knowledge of the first user preference. And then, carrying out regression analysis on the associated activity information set of the first user preference based on the basic user preference identification network to obtain current regression analysis data, determining the current network cost of the basic user preference identification network based on the current regression analysis data and the priori knowledge of the second user preference, and optimizing the network model variable of the basic user preference identification network based on the current network cost to obtain the second user preference identification network.
When the basic user preference identification network is debugged, in order to avoid the problem that the network generalization performance of the second user preference identification network is reduced, a second debugging requirement can be configured in advance, and the second debugging requirement can be understood as a debugging requirement aiming at the second user preference identification network when the second user preference identification network is debugged. Such as: the second debug requirement may be that the current network cost is less than the second stability determination value. On the basis of determining that the current network cost of the second user preference identification network is smaller than the second stability judgment value, the debugging of the base user preference identification network can be ended, so that the current regression analysis data and the priori knowledge of the second user preference are not completely consistent, and the problem that the network generalization performance of the second user preference identification network is reduced is avoided.
In the embodiment of the invention, the prior knowledge of the first user preference in the basic prior knowledge is used as the first debugging instruction, so that the basic user preference identification network can be debugged based on the basic network debugging basis until the first debugging requirement comprising maintaining the network generalization index is met, and the first user preference identification network for identifying and mining the first user preference can be timely and accurately obtained. The base user preference identification network can be debugged based on the prior knowledge of the first user preference and the prior knowledge of the second user preference on the basis of the base network debugging basis until the second debugging requirement including maintaining the network generalization index is met, so that the second user preference identification network for identifying and mining the second user preference is accurately obtained, and the operation precision and the identification reliability of the second user preference identification network are improved.
For some exemplary embodiments, the NODE113 may include those described by NODE121 through NODE 122.
And the NODE121 obtains an associated activity information set corresponding to the first user preference covered by the network debugging example from the network debugging examples in the basic network debugging basis based on the prior knowledge of the first user preference, and obtains the first network debugging basis.
The prior knowledge of the first user preference may include information such as a category of the first user preference, a size and a distribution of a visual window for annotating the first user preference, and the visual window may be in different shapes. The associated activity information set of the first user preference may be obtained from internet user activity big data according to which the underlying network is commissioned based on the visual window of the first user preference, such as: the size of the visual window of the first user preference is used as the size of the associated activity information set of the first user preference, the distribution of the visual window of the first user preference is used as the distribution of the associated activity information set of the first user preference, and the associated activity information set of the first user preference can be obtained from the Internet user activity big data. The first network debugging basis can be understood as a set of associated activity information sets of first user preference items of the internet user activity big data, and the associated activity information sets of the first user preference items of the internet user activity big data can be obtained to form the first network debugging basis. Such as: and obtaining the associated activity information sets of 20 first user preference items from 10 groups of Internet user activity big data as a first network debugging basis.
NODE122 uses the prior knowledge of the second user preference in the basic prior knowledge as a second debugging instruction, debugs the basic user preference identification network based on the first network debugging basis until the second debugging requirement including maintaining a network generalization index is met, and obtains the second user preference identification network.
In an embodiment of the invention, the second commissioning indication may be understood as a priori knowledge used for commissioning the second user preference identification network, the second commissioning indication may comprise a priori knowledge of the second user preference in the underlying a priori knowledge. On the basis of obtaining the first network debugging basis and the basic user preference identification network, carrying out regression analysis on the network debugging examples in the first network debugging basis by aiming at the second user preference, obtaining current regression analysis data, determining the current network cost of the basic user preference identification network based on the current regression analysis data and the second debugging instruction, and optimizing network model variables of the basic user preference identification network based on the current network cost, thereby obtaining the second user preference identification network.
When the basic user preference identification network is debugged, in order to reduce the probability of network generalization performance degradation of the second user preference identification network, a second debugging requirement can be configured in advance, and the second debugging requirement can be understood as a debugging requirement aiming at the second user preference identification network in the process of debugging the second user preference identification network. Such as: the second debug requirement may be that the current cycle number is greater than the cycle determination value, and the debug base user preference identification network may be ended on the basis of determining that the second debug requirement is that the current cycle number of the user preference identification network is greater than the cycle determination value, so that the current regression analysis data and the prior knowledge of the second user preference are not completely consistent, and the problem that the second user preference identification network has reduced network generalization performance is improved.
In the embodiment of the invention, based on priori knowledge of the first user preference, an associated activity information set of the first user preference covered by a corresponding network debugging example is obtained from a basic network debugging basis to obtain the first network debugging basis; and then, the priori knowledge of the second user preference in the basic priori knowledge can be used as a second debugging instruction, the basic user preference identification network is debugged based on the first network debugging basis until the second debugging requirement comprising maintaining the network generalization index is met, and the second user preference identification network is obtained. In this way, by reducing the input size of the basic user preference identification network, the operation cost of the basic user preference identification network can be reduced, so that the preference identification precision is ensured, and compared with the situation that the second user preference is identified from the big data of the internet user activity, the second user preference is identified from the associated activity information set of the first user preference, and by reducing other user preference which possibly has influence, the efficient and accurate preference identification is realized.
For some exemplary embodiments, the scheme described by NODE112 may include NODE131-NODE132 as follows.
NODE131, based on the basic user preference identification network, identifies the basic network debugging basis to obtain first user preference identification information.
The first user preference identification information may be understood as user preference identification performed on the first user preference, so as to obtain first user preference identification information, where the first user preference identification information may include information such as whether the first user preference is identified, a category of the first user preference, and a regression analysis window of the first user preference based on the identification of the first user preference. Such as: the internet user activity big data in the basic network debugging basis can be target business link activity big data (e-commerce interaction link information), the first user preference is business process guiding preference, the network is identified based on the basic user preference, the target business link activity big data (e-commerce interaction link information) is identified by taking the target business link as the user preference, the existence of the target business link is determined, and distribution of the business process guiding preference in the target business link activity big data (e-commerce interaction link information) is released in a regression analysis window mode, and the like.
NODE132 optimizes network model variables of the basic user preference identification network according to that first debug reference comparison information (such as difference information) between the basic priori knowledge and the first user preference identification information is larger than a first determination value, and obtains the first user preference identification network.
Wherein, based on determining the first user preference identification information of the network commissioning example, first commissioning reference comparison information between the basic priori knowledge of the network commissioning example and the first user preference identification information of the network commissioning example may be determined, the first commissioning reference comparison information being used to reflect a difference between the basic priori knowledge and the first user preference identification information. Such as: the basic prior knowledge indicates that 2 first user preferences exist in the internet user activity big data, the first user preference identification information indicates that 3 first user preferences exist in the internet user activity big data, and wherein the 2 first user preferences are the same as the 2 first user preferences in the basic prior knowledge, then the first commissioning reference comparison information may be determined to be an additional 1 first user preference. The first debugging reference comparison information can be a numerical value or a matrix, on the basis of determining the first debugging reference comparison information, whether the first debugging reference comparison information is larger than a set first judgment value is determined, on the basis that the first debugging reference comparison information is larger than the first judgment value, the condition that the debugging does not meet the set accuracy requirement can be understood, the network model variable of the basic user preference identification network can be optimized, the difference between the basic priori knowledge and the first user preference identification information is determined again after a plurality of cycles, and further the basic user preference identification network is continuously improved; on the basis that the first debug reference comparison information is not greater than the first determination value, it can be understood that the debug has reached the requirement of setting accuracy, and the optimization of the network model variables of the underlying user preference identification network can be ended, thereby obtaining the first user preference identification network.
In the embodiment of the invention, the network can be identified based on the basic user preference, and the basic network debugging basis is identified to obtain the first user preference identification information; therefore, according to the first debugging reference comparison information between the basic priori knowledge and the first user preference identification information is larger than the first judgment value, the network model variable of the basic user preference identification network is optimized, the first user preference identification network is timely and accurately obtained, and the problem that the network generalization performance of the first user preference identification network is reduced is solved.
For some examples, the NODE113 may include NODE133 through NODE134 as follows.
NODE133, based on the basic user preference identification network, identifies the first network debugging basis and obtains second user preference identification information.
The second user preference identification information may be understood as user preference identification performed on the second user preference, so as to obtain second user preference identification information, where the second user preference identification information may include information such as whether the second user preference is identified, a category of the second user preference, and a regression analysis window of the second user preference based on the identification of the second user preference. Such as: the first network debugging is based on an associated activity information set which can be push preference, the second user preference is office software product preference, a network is identified based on basic user preference, the office software product preference is identified as the user preference by the associated activity information set of push preference, the existence of the office software product preference is determined, and distribution of the office software product preference in the associated activity information set of push preference is released in a regression analysis window mode, and the like.
And the NODE134 optimizes the network model variable of the basic user preference identification network according to the fact that second debugging reference comparison information between the basic priori knowledge and the second user preference identification information is larger than a second judgment value, so as to obtain the second user preference identification network.
Wherein, based on determining the second user preference identification information of the network commissioning example, second commissioning reference comparison information between the basic priori knowledge of the network commissioning example and the second user preference identification information of the network commissioning example may be determined, the second commissioning reference comparison information being used to reflect a difference between the basic priori knowledge and the second user preference identification information. Such as: the basic priori knowledge indicates that 3 second user preference items exist in the internet user activity big data, the second user preference identification information indicates that 4 second user preference items exist in the internet user activity big data, and wherein the 3 second user preference items are the same as the 3 second user preference items in the basic priori knowledge, then the second debug reference comparison information may be determined to be an additional 1 second user preference item. On the basis of determining the second debug reference comparison information, whether the second debug reference comparison information is larger than a set second determination value or not can be determined, on the basis that the second debug reference comparison information is larger than the second determination value, it can be understood that the debug does not meet the set accuracy requirement, network model variables of the basic user preference identification network can be optimized, multiple cycles are performed, differences between the basic priori knowledge and the second user preference identification information are determined again, and further the basic user preference identification network is continuously improved; on the basis that the second debug reference comparison information is not greater than the second determination value, it can be understood that the debug has reached the requirement of setting accuracy, and the optimization of the network model variables of the underlying user preference identification network can be ended, thereby obtaining the second user preference identification network.
In the embodiment of the invention, the network can be identified based on the basic user preference, and the first network debugging basis is identified to obtain the second user preference identification information; therefore, the network model variable of the basic user preference identification network can be optimized according to the fact that the second debugging reference comparison information between the basic priori knowledge and the second user preference identification information is larger than the second judgment value, the second user preference identification network can be timely and accurately obtained, the debugging timeliness of the second user preference identification network is improved, the problem that the network generalization performance of the second user preference identification network is reduced is solved, and the like.
For some examples, the method may further include the following.
And performing regression analysis on at least part of network debugging examples in the network debugging basis completing the change based on the expert system analysis network to obtain regression analysis data of at least part of network debugging examples in the network debugging basis completing the change aiming at the first user preference or the second user preference.
On the basis of changing the basic network debugging basis to obtain the changed network debugging basis, regression analysis can be performed on at least part of network debugging examples in the changed network debugging basis based on the expert system analysis network to obtain regression analysis data of at least part of network debugging examples in the changed network debugging basis about the first user preference or the second user preference. Such as: the network debugging basis for completing the change comprises 100 groups of Internet user activity big data and basic priori knowledge of each Internet user activity big data for completing the change, 50 groups of Internet user activity big data are arbitrarily obtained, regression analysis is carried out on the 50 groups of Internet user activity big data based on an expert system analysis network for completing the debugging, and regression analysis data of the 50 groups of Internet user activity big data, wherein the 50 groups of Internet user activity big data comprise 51 first user preference items and 79 second user preference items, are obtained. The number of at least some network commissioning instances in the network commissioning basis that completed the change may be consistent or may differ from the number of at least some network commissioning instances in the underlying network commissioning basis.
And the NODE142 ends the improvement of the expert system analysis network on the basis that the regression analysis data of the network debugging example of the set duty ratio in the network debugging basis for completing the change is not in the target prediction category.
Wherein, based on obtaining regression analysis data of at least part of the network debugging examples in the network debugging bases for completing the change about the first user preference or the second user preference, the regression analysis data of at least part of the network debugging examples in the network debugging bases for completing the change about the first user preference or the second user preference can be compared with basic prior knowledge for completing the change, and knowledge of the current circulation times can be determined to be endowed with quality information. Knowledge of the current number of loops gives quality information that may reflect whether the regression analysis data for the first user preference or the second user preference is of the target prediction category for the portion of the network commissioning example. The set duty ratio (e.g., 0.8) of the first user preference item and the second user preference item for the target prediction category may be configured in advance, and the accuracy requirement meeting the setting is determined based on determining that the duty ratio of the regression analysis data for the target prediction category is greater than the set duty ratio (e.g., 0.8). Such as: regression analysis data regarding the first user preference or the second user preference for at least some of the network commissioning bases that determine the completed change may be an additional 1 first user preference, an additional 10 second user preferences. The first user preference is not the target prediction category, the 9 second user preference is not the target prediction category, the 1 second user preference is the target prediction category, the first user preference is not the target prediction category with the duty ratio of 100%, the second user preference is not the target prediction category with the duty ratio of 0.9, then the accuracy of the current expert system analysis network can be considered to meet the set condition, and the improvement of the expert system analysis network can be finished.
For some examples, on the basis of regression analysis data of the network commissioning example according to the set duty ratio in the network commissioning basis completing the change as the target prediction category, the expert system analysis network may be continuously improved by a number of loops until the set accuracy requirement can be met.
In the embodiment of the invention, the improvement of the expert system analysis network can be finished in time on the basis that the regression analysis data of the preset number of network debugging examples in the network debugging basis for completing the change is not the target prediction type, so that the problem that the recognition accuracy of the expert system analysis network is difficult to effectively improve due to multiple improvements is avoided.
Another embodiment of the present invention provides an artificial intelligence based user profile analysis method comprising NODE201 through NODE205 as follows.
NODE201, obtains expert system analysis network for completing debugging and uses as basic network debugging basis for debugging the expert system analysis network.
And performing regression analysis on at least part of network debugging examples in the basic network debugging basis based on the expert system analysis network to obtain regression analysis data of the part of network debugging examples aiming at the first user preference or the second user preference.
The NODE201-NODE202 correspond to the NODE101-NODE102 described above, respectively.
NODE203, carrying out joint analysis on regression analysis data of the partial network debugging examples and target priori knowledge of the corresponding network debugging examples in the basic network debugging basis to obtain debugging reference comparison information.
The debug reference comparison information may reflect whether regression analysis data of a part of network debug examples is adapted to target priori knowledge of a corresponding network debug example in the basic network debug basis, and since the first user preference and the second user preference in the basic priori knowledge are annotated in a set manner, the basic priori knowledge may also be referred to as target priori knowledge. It may be determined whether the first user preference in the regression analysis data matches the first user preference in the target prior knowledge based on information such as a distribution of the first user preference or the second user preference. Such as: the regression analysis data may be that there are 3 first user preference items, the distribution feature variables of the regression analysis window of the first user preference items are location1, location2 and location3, respectively, the target priori knowledge is that there are 2 first user preference items, and the distribution feature variables of the visual window of the first user preference items are location1 and location2, respectively, so that it may be determined that 2 first user preference items in the regression analysis data adapt to corresponding target priori knowledge, and 1 first user preference item does not adapt.
NODE204, according to the regression analysis data of the partial network debugging examples, not adapting to the target priori knowledge of the corresponding network debugging examples in the basic network debugging basis, and outputting the regression analysis data of the partial network debugging examples and the target priori knowledge of the corresponding network debugging examples to obtain derivative evaluation information.
Wherein the derived evaluation information is used for reflecting whether regression analysis data of the user preference for which the debug reference comparison information is not adapted in the partial network debug example is a target prediction category, and since the knowledge-imparting quality information reflects whether regression analysis data of the partial network debug example with respect to the first user preference or the second user preference is a target prediction category, the derived evaluation information can be understood as knowledge-imparting quality information based on the manual collation result. On the basis of determining that regression analysis data of part of network debugging examples are not suitable for target priori knowledge of the network debugging examples, the comparison debugging reference comparison information can be compared with information such as states, sizes, distribution and the like between unsuitable user preference items and user preference items in basic priori knowledge through manual checking rules, so that derivative evaluation information is obtained. Such as: determining knowledge-imparting quality information includes regression analysis data in which 2 first user preferences fit corresponding target prior knowledge, 1 first user preference does not fit, and the first user preference is an actual/true first user preference, i.e., is not a target prediction category.
NODE205 gives quality information to change the basic network debugging basis based on the knowledge, and obtains the network debugging basis for completing the change so as to improve the expert system analysis network.
The NODE205 corresponds to the NODE104 described above.
In the embodiment of the invention, the debugging reference comparison information is obtained by carrying out joint analysis on regression analysis data of part of network debugging examples and target priori knowledge of corresponding network debugging examples in a basic network debugging basis; the debugging reference comparison information reflects whether regression analysis data of part of network debugging examples are matched with target priori knowledge of corresponding network debugging examples in the basic network debugging basis; in this way, the regression analysis data of the part of network debugging examples and the target priori knowledge of the corresponding network debugging examples in the basic network debugging basis can be output according to the target priori knowledge of the corresponding network debugging examples in the regression analysis data of the part of network debugging examples so as to obtain derivative evaluation information; the derived evaluation information is used for reflecting whether regression analysis data corresponding to the network debugging examples in the partial network debugging examples is of a target prediction type, so that the derived evaluation information can be conveniently and accurately determined in time.
Furthermore, still another embodiment of the present invention provides an artificial intelligence based user profile analysis method including the following NODE301 to NODE305.
NODE301, obtains an expert system analysis network to complete the debugging and serves as a basis for debugging the expert system analysis network.
And performing regression analysis on at least part of network debugging examples in the basic network debugging basis based on the expert system analysis network to obtain regression analysis data of the part of network debugging examples aiming at the first user preference or the second user preference.
And NODE303, combining regression analysis data of the partial network debugging examples and basic priori knowledge of the corresponding network debugging examples in the basic network debugging basis, and determining knowledge-endowed quality information.
Wherein the knowledge-endowed quality information reflects whether regression analysis data of the portion of the network commissioning example for the first user preference or the second user preference is of a target prediction class.
NODE301-NODE303 correspond to NODE101-NODE103, respectively, described above.
NODE304, according to the knowledge endowing quality information, reflecting that the regression analysis data of the part of network debugging examples are not the target prediction category, and changing the regression analysis data of the part of network debugging examples into the prior knowledge of the corresponding network debugging examples.
On the basis that the knowledge endowing quality information is determined to reflect that regression analysis data of part of network debugging examples are not in the target prediction category, the regression analysis data of the network debugging examples which are not in the target prediction category can be changed into priori knowledge of the corresponding network debugging examples, so that basic priori knowledge of the completed change is obtained, and the method can be used for further improving expert system analysis networks. Such as: determining knowledge-endowed quality information comprises regression analysis data comprising basic prior knowledge corresponding to 2 first user preference adaptations, 1 first user preference is not adapted, and the first user preference is an actual first user preference and is not a target prediction category. Then, the regression analysis data for the first user preference and the list of mappings between the regression analysis data and the network debug instance may be added to the basic prior knowledge to obtain the basic prior knowledge of the completed change.
NODE305 uses regression analysis data of the partial network debug instance as the target prediction category without changing prior knowledge of the corresponding network debug instance.
The knowledge endowed quality information is determined to reflect the prior knowledge of the corresponding network debugging example on the basis of the regression analysis data of the partial network debugging example serving as the target prediction type, and the improved expert system analysis network and the like can be ended on the basis of the regression analysis data of the partial network debugging example serving as the target prediction type. Such as: determining knowledge-imparting quality information includes regression analysis data including basic prior knowledge that 2 first user preferences fit, 1 first user preference does not fit, and the first user preference is not a true first user preference, such as for a target prediction category. The current basic priori knowledge, that is, the accuracy of the current basic priori knowledge is higher, can not be changed, so that the recognition accuracy of the current expert system analysis network is determined to meet the requirement.
In the embodiment of the invention, according to the knowledge endowing quality information, the regression analysis data of part of network debugging examples are reflected to be not in a target prediction category, and the regression analysis data of part of network debugging examples are changed into priori knowledge of the corresponding network debugging examples; according to the regression analysis data of the part of network debugging examples, the prior knowledge of the corresponding network debugging examples is unchanged, so that the prior knowledge can be changed accurately and timely, the expert system analysis network can be further improved, or the improvement of the expert system analysis network can be finished timely, and the like.
When the annotation information quantity of the user preference items in the Internet user activity big data reaches a certain scale and the priori knowledge of the user preference items does not have more deviation, the basic user preference identification network can be debugged to obtain the expert system analysis network with better precision based on the Internet user activity big data and the priori knowledge. That is, some incorrect prior knowledge can be used as interference information, so that debugging of the expert system analysis network is not affected, and the expert system analysis network has certain resistance to the disturbance information.
Compared with the thought of repeatedly modifying the priori knowledge of the basic network debugging basis back and forth, the embodiment of the invention can reduce resource consumption, obtain the network debugging basis for completing the change with better accuracy, and improve the identification precision and the reliability of the expert system analysis network.
In some independent embodiments, after inputting the user activity big data to be analyzed into the finishing improved expert system analysis network to obtain the user preference analysis result output by the finishing improved expert system analysis network, the method may further include the following: determining information pushing requirements for big data of user activities to be analyzed based on the analysis result of the user preference; and customizing a pushing strategy by utilizing the information pushing requirement, and applying the pushing strategy.
For example, if the user preference analysis result includes push preference of the user, information push requirements of big data of the user activity to be analyzed can be further deeply mined, so that a push strategy is customized in a targeted and personalized manner, and accurate and efficient data push processing is performed by using the push strategy.
In some independent embodiments, determining information push requirements for user activity big data to be analyzed based on the user preference analysis results may include the following: acquiring a plurality of first information push demand data and a plurality of second information push demand data, wherein the first information push demand data is information push demand data carrying a mark, and the second information push demand data is personalized push demand data not carrying a mark; based on the identification carried by each first information push demand data, extracting an identification indication field (demand feature vector) corresponding to each first information push demand data, and determining personalized push demand data in the plurality of first information push demand data according to the identification indication field; generating a personalized pushing demand sample set according to the determined personalized pushing demand data and the second information pushing demand data, and calculating the association degree between each information pushing demand data in the personalized pushing demand sample set and pushing preference data corresponding to the analysis result of the user preference item to obtain a plurality of calculation results; if it is determined that information push demand data meeting association requirements exists in the personalized push demand sample set according to the multiple calculation results, the push preference data corresponding to the user preference analysis results are determined to be personalized push demand data, and information push demands are determined based on personalized demand keywords corresponding to the push preference data corresponding to the user preference analysis results. In this way, personalized analysis can be performed on the push preference data corresponding to the analysis result of the user preference by using the information push demand data with different identifications, so that the differentiation determination of the information push demand is realized, and the feature recognition degree and quality of the information push demand are improved.
In some independent embodiments, determining personalized push demand data in the plurality of first information push demand data according to the identification indication field includes: inputting identification description variables corresponding to the first information pushing demand data into a configured demand data processing network to obtain personalized scores corresponding to the first information pushing demand data output by the demand data processing network, wherein the personalized scores are the possibility that the first information pushing demand data is personalized pushing demand data, the demand data processing network is configured based on a plurality of configuration information pushing demand data carrying the identification and information pushing demand data categories corresponding to the configuration information pushing demand data, and the information pushing demand data categories comprise personalized pushing demand data and non-personalized pushing demand data; and determining personalized pushing demand data in the plurality of first information pushing demand data according to the personalized scores corresponding to the first information pushing demand data. In this way, the personalized pushing demand data is determined by introducing the demand data processing network (classification network), so that the accuracy and the credibility of the determination of the personalized pushing demand data can be ensured.
Based on the same or similar inventive concept, please refer to fig. 2, there is also provided a schematic architecture diagram of an application environment 30 of an artificial intelligence based user portrait analysis method, which includes an artificial intelligence based user portrait analysis system 10 and a service user terminal 20 that communicate with each other, where the artificial intelligence based user portrait analysis system 10 and the service user terminal 20 implement or partially implement the technical scheme described in the above method embodiments at runtime.
Further, there is also provided a readable storage medium having stored thereon a program which when executed by a processor implements the above-described method.
In the embodiments provided in the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus and method embodiments described above are merely illustrative, for example, flow diagrams and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules in the embodiments of the present invention may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a network device, or the like) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (4)
1. An artificial intelligence-based user portrayal analysis method, which is applied to an artificial intelligence-based user portrayal analysis system, comprises the following steps:
when a portrait analysis request is received, acquiring big data of user activities to be analyzed based on the portrait analysis request;
inputting the big data of the user activities to be analyzed into an expert system analysis network for finishing improvement, and obtaining a user preference analysis result output by the expert system analysis network for finishing improvement;
the expert system analysis network for accomplishing the improvement is improved by the following steps:
obtaining an expert system analysis network for completing debugging and a basic network debugging basis for debugging the expert system analysis network; the network debugging examples in the basic network debugging foundation are Internet user activity big data containing first user preference items and second user preference items, the associated activity information sets of the second user preference items are in the associated activity information sets of the first user preference items, the basic prior knowledge of at least part of the network debugging examples in the basic network debugging foundation does not realize comprehensive annotation of the first user preference items and/or the second user preference items, the associated activity information sets of the first user preference items are content sets containing the first user preference items in the Internet user activity big data, and the associated activity information sets of the second user preference items are content sets containing the second user preference items in the Internet user activity big data;
Performing regression analysis on at least part of network debugging examples in the basic network debugging basis based on the expert system analysis network to obtain regression analysis data of the part of network debugging examples aiming at the first user preference or the second user preference;
combining regression analysis data of the partial network debugging examples and basic priori knowledge of the corresponding network debugging examples in the basic network debugging basis to determine knowledge endowed quality information; wherein the knowledge-endowed quality information reflects whether regression analysis data of the partial network commissioning example for the first user preference or the second user preference is of a target prediction class;
changing the basic network debugging basis based on the knowledge endowed quality information to obtain a network debugging basis for completing the change so as to improve the expert system analysis network;
the first user preference and/or the second user preference in the basic prior knowledge are annotated by a set way; the expert system analysis network includes a first user preference identification network and a second user preference identification network; the method further comprises the steps of:
Obtaining an unbugged underlying user preference identification network;
taking the priori knowledge of the first user preference in the basic priori knowledge as a first debugging instruction, and debugging the basic user preference identification network based on the basic network debugging basis until the first debugging requirement including maintaining network generalization indexes is met, so as to obtain the first user preference identification network;
based on the prior knowledge of the first user preference and the prior knowledge of the second user preference, debugging the basic user preference identification network based on the basic network debugging basis until the second debugging requirement including maintaining network generalization indexes is met, and obtaining the second user preference identification network;
the step of debugging the basic user preference identification network based on the prior knowledge of the first user preference and the prior knowledge of the second user preference until meeting a second debugging requirement including maintaining a network generalization index, to obtain the second user preference identification network, includes:
based on the prior knowledge of the first user preference, obtaining an associated activity information set of the first user preference covered by a corresponding network debugging example from the network debugging examples in the basic network debugging basis to obtain a first network debugging basis;
Taking the priori knowledge of a second user preference in the basic priori knowledge as a second debugging instruction, and debugging the basic user preference identification network based on the first network debugging basis until the second debugging requirement comprising maintaining a network generalization index is met, so as to obtain the second user preference identification network;
the determining knowledge endowing quality information by combining regression analysis data of the partial network debugging examples and basic priori knowledge of corresponding network debugging examples in the basic network debugging basis comprises the following steps:
performing joint analysis on regression analysis data of the partial network debugging examples and target priori knowledge of the corresponding network debugging examples in the basic network debugging basis to obtain debugging reference comparison information; the debugging reference comparison information reflects whether regression analysis data of the partial network debugging examples are suitable for target priori knowledge of corresponding network debugging examples in the basic network debugging basis;
according to the regression analysis data of the partial network debugging examples, not adapting to the target priori knowledge of the corresponding network debugging examples in the basic network debugging basis, outputting the regression analysis data of the partial network debugging examples and the target priori knowledge of the corresponding network debugging examples to obtain derivative evaluation information; the derived evaluation information is used for reflecting whether regression analysis data of the corresponding network debugging examples in the partial network debugging examples is the target prediction category or not;
The step of debugging the basic user preference identification network based on the basic network debugging basis until the first debugging requirement including maintaining network generalization indexes is met, and obtaining the first user preference identification network comprises the following steps: identifying the basic network debugging basis based on the basic user preference identification network to obtain first user preference identification information; according to the first comparison data between the basic priori knowledge and the first user preference identification information does not accord with a first requirement, optimizing network model variables of the basic user preference identification network to obtain the first user preference identification network;
the step of debugging the basic user preference identification network based on the basic network debugging basis until the basic user preference identification network meets a second debugging requirement including maintaining network generalization indexes, and obtaining the second user preference identification network comprises the following steps: identifying the first network debugging basis based on the basic user preference identification network to obtain second user preference identification information; according to the fact that second comparison data between the basic priori knowledge and the second user preference identification information does not meet second requirements, network model variables of the basic user preference identification network are optimized, and the second user preference identification network is obtained;
The method further comprises the steps of:
performing regression analysis on at least part of network debugging examples in the network debugging basis completing the change based on the expert system analysis network to obtain regression analysis data of at least part of network debugging examples in the network debugging basis completing the change aiming at the first user preference or the second user preference;
and finishing the improvement of the expert system analysis network on the basis that regression analysis data of the network debugging examples with the set duty ratio in the changed network debugging basis is not in the target prediction category.
2. The method according to claim 1, wherein said altering the base network debug evidence based on the knowledge-based quality information to obtain an altered network debug evidence comprises:
according to the knowledge endowing quality information, reflecting that regression analysis data of the partial network debugging examples are not of the target prediction category, and changing the regression analysis data of the partial network debugging examples into priori knowledge of the corresponding network debugging examples;
and according to regression analysis data of the partial network debugging examples, the target prediction category is obtained, and the prior knowledge of the corresponding network debugging examples is unchanged.
3. An artificial intelligence based user portrayal analysis system comprising a processor and a memory; the processor is communicatively connected to the memory, and the processor is configured to read the computer program from the memory and execute the computer program to implement the method of claim 1 or 2.
4. A computer readable storage medium, characterized in that a program is stored thereon, which program, when being executed by a processor, implements the method of claim 1 or 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211508615.6A CN115905698B (en) | 2022-11-29 | 2022-11-29 | User portrait analysis method and system based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211508615.6A CN115905698B (en) | 2022-11-29 | 2022-11-29 | User portrait analysis method and system based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115905698A CN115905698A (en) | 2023-04-04 |
| CN115905698B true CN115905698B (en) | 2023-12-26 |
Family
ID=86472588
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211508615.6A Active CN115905698B (en) | 2022-11-29 | 2022-11-29 | User portrait analysis method and system based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115905698B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20160075103A (en) * | 2014-12-19 | 2016-06-29 | 연세대학교 산학협력단 | Method and Device for Predicting Preference of New Content |
| CN112837087A (en) * | 2020-12-16 | 2021-05-25 | 北京交通大学 | User portrait construction method oriented to consultation service system |
| CN113656649A (en) * | 2021-08-13 | 2021-11-16 | 南方电网数字电网研究院有限公司 | Generation and storage algorithm and system for label portrait data |
| CN114399352A (en) * | 2021-12-22 | 2022-04-26 | 中国电信股份有限公司 | Information recommendation method and device, electronic equipment and storage medium |
| CN115203534A (en) * | 2022-06-15 | 2022-10-18 | 丁伟俊 | Activity preference analysis method and system applied to digital space |
-
2022
- 2022-11-29 CN CN202211508615.6A patent/CN115905698B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20160075103A (en) * | 2014-12-19 | 2016-06-29 | 연세대학교 산학협력단 | Method and Device for Predicting Preference of New Content |
| CN112837087A (en) * | 2020-12-16 | 2021-05-25 | 北京交通大学 | User portrait construction method oriented to consultation service system |
| CN113656649A (en) * | 2021-08-13 | 2021-11-16 | 南方电网数字电网研究院有限公司 | Generation and storage algorithm and system for label portrait data |
| CN114399352A (en) * | 2021-12-22 | 2022-04-26 | 中国电信股份有限公司 | Information recommendation method and device, electronic equipment and storage medium |
| CN115203534A (en) * | 2022-06-15 | 2022-10-18 | 丁伟俊 | Activity preference analysis method and system applied to digital space |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115905698A (en) | 2023-04-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101788907B (en) | Automatic generation method for combination service simplified test case and device therefor | |
| CN107391368B (en) | Method for analyzing consistency of source code and target code in airborne software | |
| US20230108309A1 (en) | Methods and systems for gas meter replacement prompt based on a smart gas internet of things | |
| CN113608882A (en) | Information processing method and system based on artificial intelligence and big data and cloud platform | |
| CN110287104A (en) | Method for generating test case, device, terminal and computer readable storage medium | |
| CN113360763A (en) | Service attention tendency prediction method based on artificial intelligence and artificial intelligence cloud system | |
| CN109800776A (en) | Material mask method, device, terminal and computer readable storage medium | |
| CN114637268A (en) | Industrial production control method and system based on artificial intelligence and cloud platform | |
| CN115905698B (en) | User portrait analysis method and system based on artificial intelligence | |
| CN115511233A (en) | Supply chain process reproduction method and system based on process mining | |
| CN110766086B (en) | Method and device for fusing multiple classification models based on reinforcement learning model | |
| KR20220041600A (en) | Method of evaluating quality of smart factory data | |
| Khalid et al. | Early cost estimation of software reworks using fuzzy requirement-based model | |
| CN117035065A (en) | Model evaluation method and related device | |
| CN112529218A (en) | Building safety detection method and system based on correlation analysis | |
| CN115563069B (en) | Data sharing processing method and system based on artificial intelligence and cloud platform | |
| CN115629995B (en) | Software defect positioning method, system and equipment based on multi-dependency LSTM | |
| CN111859862A (en) | Text data labeling method and device, storage medium and electronic device | |
| CN104123469A (en) | Detection scheduling system and method for context consistency in pervasive computing environment | |
| CN113239272B (en) | Intention prediction method and intention prediction device of network management and control system | |
| CN110597874B (en) | Data analysis model creation method and device, computer equipment and storage medium | |
| CN116561289B (en) | Real-time monitoring method and system for digital factory | |
| CN111897535A (en) | Grammar error correction method, device, computer system and readable storage medium | |
| Czwick et al. | Data Pre-Processing Method for Industrie 4.0 Applications | |
| CN115730283B (en) | Account login wind control system, method, computer equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20231206 Address after: 550000 room 4401, 44F, unit 1, building 15, Zone V, Wulichong flower orchard project, Wulichong sub district office, Nanming District, Guiyang City, Guizhou Province [Wulichong office] Applicant after: Guizhou Youte Shulian Technology Co.,Ltd. Address before: No. 89 Yanhe Street, Erdao District, Changchun City, Jilin Province, 130000 Applicant before: Liu He |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |