Advertisement recommendation effect evaluation system based on multi-dimensional user portraits

Abstract

The invention discloses an advertisement recommendation effect evaluation system based on multi-dimensional user portraits, and relates to the technical field of artificial intelligence, functional modules of the system comprise a user portrait quantification module, a space-time correction module, a sample weight correction module, a joint evaluation score calculation module and a normalization quantification output module. According to the method, a user portrait abnormal risk measurement mechanism based on space-time correction is set, normalized modeling is carried out in combination with the sample weight, a unified abnormal risk combined evaluation score is formed, the defect of abnormal risk isolated evaluation in the prior art is effectively overcome, and the stability and credibility of advertisement recommendation effect evaluation are improved. According to the technical means, a solid data basis is provided for risk identification and effect quantification of the recommendation model, and an advertisement recommendation system is promoted to develop towards a more accurate and intelligent direction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an advertisement recommendation effect evaluation system based on multi-dimensional user portrait. BACKGROUND

[0002] With the popularity of the Internet and mobile terminals, the advertisement recommendation system plays an important role in digital marketing. Accurate evaluation of the effect of advertisement recommendation is not only directly related to the rational allocation of marketing resources, but also affects user experience and platform revenue. Traditional advertisement recommendation effect evaluation relies on single-dimensional user behavior data or simple statistical indicators, which is difficult to fully reflect the changes in user interest and the actual impact of the advertisement. In recent years, multi-dimensional user portrait technology has been gradually applied to advertisement recommendation, which attempts to achieve more accurate user interest description and recommendation effect judgment by integrating user behavior data, geographic location, time characteristics and other multiple dimensions. In the prior art, although some methods attempt to introduce user portrait spatiotemporal information for recommendation risk evaluation, there is a common problem of evaluating sample abnormal scores separately, lacking joint modeling of sample weights, resulting in insufficient robustness and accuracy of the overall evaluation. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides an advertisement recommendation effect evaluation system based on multi-dimensional user portrait to solve the problems raised in the background art.

[0004] To achieve the above purpose, the present application provides the following technical scheme: In a first aspect, the present application provides an advertisement recommendation effect evaluation system based on multi-dimensional user portrait, the functional modules of the system include: user portrait quantification module, spatiotemporal correction module, sample weight correction module, joint evaluation score calculation module and normalized quantization output module. The functional modules of the system include the following steps: Based on the user portrait quantification module, the attributes of the collected user portrait are quantified to obtain a portrait vector; The spatiotemporal correction module is used to correct the portrait vector obtained by the user portrait quantification module in space and time to obtain a spatiotemporal correction portrait vector; The sample weight correction module is used to correct the spatiotemporal correction portrait vector output by the spatiotemporal correction module to obtain a sample weight; Based on the sample weight output by the sample weight correction module, the joint evaluation score calculation module is used for joint evaluation to obtain a joint evaluation score; The joint evaluation score output by the joint evaluation score calculation module is input into a normalization quantization output module to perform normalization quantization to obtain a quantization output score as a basis for measuring the size of the recommendation risk in the advertisement recommendation loop. The higher the score, the higher the advertisement recommendation risk.

[0005] Further optimization of the technical solution, the input portrait in the user portrait quantization module contains: user basic information, user interest label, user behavior data and user geographic information, through One-Hot coding technology, WordEmbedding technology and normalization processing to quantize the above portrait information, form a portrait vector .

[0006] Further optimization of the technical solution, the space-time correction module first performs space-time weight modeling, then performs space-time deviation correction, and finally performs corrected portrait reconstruction; In the space-time weight modeling, a space-time weight is applied to each portrait component, and the space-time weight is determined by a space-time factor. The corresponding space-time weight modeling is a space-time function, and the formula is: ; Among them, : space-time weight; : collection time; : collection location; : Sigmoid activation function; : sensitivity of space-time weight to time; : sensitivity of space-time weight to location; : bias, to ensure stable weight distribution.

[0007] Further optimization of the technical solution, the space-time correction module eliminates the deviation of the portrait caused by the space-time condition when performing space-time deviation correction, and realizes it by correcting each component. The corresponding model formula is: ; Among them, : the corrected portrait component; : the first portrait component before correction; : the first portrait component before correction; : space-time weight, multiply each component for correction.

[0008] ​Further optimization of the technical solution, the space-time correction module after correction, the corresponding reconstruction formula is: Among them, : New image vector formed after space-time correction.

[0009] Further optimization of the technical solution, the sample weight correction module first performs sample weight modeling, then performs sample weight correction, and finally outputs the weight, the sample The anomaly score of the sample is: Among them, : The space-time correction image vector of the i-th sample, : The balanced vector obtained by the overall mean of the space-time correction image, , wherein i is the total number of samples.

[0010] Further optimization of the technical solution, the sample weight correction module assigns a weight to each sample according to the anomaly score when performing sample weight correction: Among them, : Sample weight; : Anomaly score of the sample : Weight decay factor.

[0011] Further optimization of the technical solution, the sample weight correction module finally outputs the weight of each sample: And thus form the sample weight: .

[0012] Further optimization of the technical solution, the joint evaluation score calculation module first performs weight normalization, then performs joint evaluation score modeling, and obtains the joint evaluation score .

[0013] Secondly, the embodiment of the application provides a computer device, including a memory and a processor, the memory stores a computer program, wherein: the computer program instructions are executed by the processor to realize the steps of the advertisement recommendation effect evaluation system based on the multi-dimensional user portrait of the first aspect of the application.

[0014] ​​​​​​​In a third aspect, the embodiments of the present application provide a computer readable storage medium, which stores a computer program, wherein the computer program instructs a processor to implement the steps of the system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait according to the first aspect of the present application.

[0015] Compared with the prior art, the present application provides a system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait, which has the following beneficial effects: The system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait effectively overcomes the deficiency of isolated evaluation of abnormal risk in the prior art by setting a user portrait abnormal risk measurement mechanism based on space-time correction and jointly normalizing modeling with sample weight to form a unified abnormal risk joint evaluation score, thereby improving the stability and credibility of the evaluation of the effect of advertisement recommendation. The technical means provides a solid data foundation for risk identification and effect quantification of the recommendation model and promotes the development of the advertisement recommendation system in a more accurate and intelligent direction. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 A flowchart of the system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait is provided. Figure 2 A space-time correction flowchart of the system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait is provided. Figure 3 A sample weight correction flowchart of the system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait is provided. Figure 4 A joint evaluation flowchart of the system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait is provided. Figure 5 A functional module diagram of the system for evaluating the effect of advertisement recommendation based on multi-dimensional user portrait is provided. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0019] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be appreciated that the present application can be practiced in a variety of ways beyond the specific details set forth herein, assuming that the fundamental underlying principles are maintained. It should also be appreciated that the present application can be practiced by employing only some of the described implementations. Accordingly, the present application is not to be limited in scope by the specific implementations disclosed in the description, which are intended as illustrations of one or more aspects of the present application. Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations.

[0020] Second, the "one embodiment" or "an embodiment" referred to herein means a specific feature, structure, or characteristic under discussion that can be included in at least one implementation of the present application. The "in one embodiment" appearing in various places in the specification does not all refer to the same embodiment, nor is it an embodiment that is separate or mutually exclusive with other embodiments.

[0021] Embodiment I: With reference to Figures 1-4 For the first embodiment of the present application, the embodiment provides an advertisement recommendation effect evaluation system based on multi-dimensional user portrait, comprising the following steps: S1, using the collected user portrait to quantize the attribute, and obtaining the portrait vector; The multi-dimensional user portrait data input in step S1 is mainly from the behavior collection devices of users on mobile terminals, PC terminals, Internet of Things devices and other terminals, including but not limited to application clients, web browsers, positioning sensors and social platform interfaces. After the multi-dimensional information such as click behavior, browsing records, geographic position, time stamp and interest tags of the user are collected uniformly through data collection technology, the original data is denoised, format converted and feature extracted by using data cleaning and preprocessing technology, and finally the multi-dimensional discrete and continuous attributes are converted into structured user portrait vectors through numerical coding and vectorization technology, providing standardized and calculable portrait input for subsequent steps.

[0022] The input portrait of step S1 includes: User basic information, including age, gender and other information tags; User interest tags, including sports, movies, games and other category tags; User behavior data, including browsing time, click times and other numerical information; User geographic information, i.e. location information.

[0023] Step S1 uses One-Hot coding technology to numerically encode the classification attributes such as "gender" and "region" based on the above portrait information. The gender is "male" and "female", which is encoded as "[1, 0]" and "[0, 1]" by One-Hot coding, and the region information is One-Hot coded according to the administrative region.

[0024] The Word Embedding technology is used to embed each interest label. If a user has multiple labels, the final interest embedding vector is obtained by vector operation on each label: ; For user behavior data, the numerical information such as browsing time and click times is normalized to form a decimal number in the domain [0, 1] to eliminate dimensional differences.

[0025] Finally, these quantified information is concatenated to form the portrait vector .

[0026] S2, using the portrait vector to perform space-time correction to obtain a space-time corrected portrait vector; Step S2 performs space-time correction operation on the portrait vector obtained in step S1 , including space-time weight modeling, space-time bias correction and corrected portrait reconstruction.

[0027] Step S2 first performs space-time weight modeling, which applies space-time weight to each portrait component. The space-time weight is determined by the space-time factor. The corresponding space-time weight modeling is a space-time function, and its formula is: ; Where, : space-time weight; : collection time; : collection location; : Sigmoid activation function; : sensitivity of space-time weight to time, which is obtained by training historical samples through existing data fitting technology. This process includes regression analysis of the relationship between space-time conditions and recommendation risk at different times, from which the value that best measures the effect of time on space-time weight is determined; : sensitivity of space-time weight to location, which is determined by regression analysis of risk samples in different geographic locations to determine the optimal fitting parameter when location affects space-time weight; : bias, which ensures stable weight distribution and is determined by existing data verification technology. When the fitted space-time weight is verified, numerical correction is performed to ensure stable weight distribution. This verification technology usually detects bias in training samples. If there is a overall shift, adjust to eliminate bias.

[0028] Step S2 then performs spatiotemporal deviation correction to eliminate the offset of the image caused by spatiotemporal conditions. This is achieved by correcting each component. The corresponding model formula is: ; in, : After correction A portrait component; : No. before correction A portrait component; : spatiotemporal weights, multiplicative correction for each component.

[0029] Finally, step S2 reconstructs the corrected image. The corresponding reconstruction formula is: ; in, : The new image vector formed after spatiotemporal correction.

[0030] The innovation of the spatiotemporal correction process in step S3 lies in the integrated data correction process of joint spatiotemporal weight modeling, spatiotemporal deviation correction, and post-correction portrait reconstruction. This overcomes the data insufficiency of the existing technology, which only performs simple normalization and lacks weight correction for each component. This makes the spatiotemporal correction process more adaptable to data under different spatiotemporal conditions. Specifically, this process implements intelligent data correction by assigning a weight dependent on spatiotemporal conditions to each component. This correction method significantly reduces the interference of spatiotemporal conditions on the credibility of portrait data, providing an accurate and robust data foundation for evaluating recommendation effects in subsequent steps.

[0031] S3. Use the spatiotemporal correction portrait vector to perform sample weight correction to obtain sample weight; Step S3 performs sample weight correction on the spatiotemporal correction image vector outputted from step S2 to adapt the effect of different samples on the overall training process.

[0032] Step S3 first performs sample weight modeling. In order to measure the credibility of each sample for model training, it is necessary to perform weight modeling on the spatiotemporal correction portrait. The weight depends on the degree of abnormality of the sample in the spatiotemporal correction portrait. The degree of abnormality can be measured by the Euclidean distance to measure the deviation of the sample from the overall mean value.

[0033] The corresponding model formula is: ; in, :sample Anomaly score : the spatio-temporal correction image vector of the first sample; : the equalization vector obtained by the overall mean of the spatio-temporal correction image, , wherein is the total number of samples.

[0034] Step S3 then proceeds with sample weight correction, assigning a weight to each sample according to the anomaly score: ; wherein : sample weight; : anomaly score of the sample The anomaly score of the sample is quantified by measuring the difference between the spatio-temporal correction image vector of each sample and the overall mean. This process is calculated by using the Euclidean distance, i.e. the Euclidean distance value between each sample and the overall mean is the anomaly score of the sample; : weight decay factor, a normal number, determines the rate of weight decline as the anomaly score increases. This weight decay factor is determined by modeling the anomaly score of each sample in the spatio-temporal correction image sample. In this process, first, the difference between the spatio-temporal correction image vector and the overall mean is measured to obtain the anomaly score sequence. Then, the anomaly score is statistically analyzed to obtain the data center and measure the distribution dispersion when the anomaly score deviates significantly. Finally, the weight decay factor is adaptively set according to the dispersion of the anomaly score. This setting ensures that the weight decreases steadily and moderately as the anomaly score increases, thereby suppressing the interference of abnormal samples on the overall training process, while avoiding unnecessary inhibition of normal samples. This process realizes data adaptation, robustness and stability in weight decay setting, and is suitable for data distribution under different conditions, providing a more robust and reliable data basis for the overall model.

[0035] Step S3 finally outputs the weight of each sample: ; and thus forms the sample weight: .

[0036] ​The innovation of step S3 mainly lies in jointly correcting the spatiotemporal profile and the sample anomaly score to model the sample weight. The joint modeling overcomes the problem of insufficient robustness caused by the lack of differentiated modeling in the prior art, which only gives a uniform weight to the sample, and provides a more stable and reliable data basis for the subsequent link. The prior art usually gives each sample an approximately same weight, or sets the weight according to the sample label only, lacking quantitative modeling of the actual data distribution information of the sample. The present method relies on the spatiotemporal profile to measure the anomaly score of each sample, and then corrects the weight according to the anomaly score, thereby significantly suppressing the adverse effect of abnormal samples on the overall model training. This sample weight correction method has more applicable conditions and stronger robustness, and provides a more stable and reliable data basis for the subsequent link.

[0037] S4, using the sample weight for joint evaluation to obtain a joint evaluation score; Step S4 evaluates each sample by using the sample weight information, and the joint evaluation link includes weight normalization, joint evaluation score modeling, and joint evaluation score output.

[0038] Step S4 first performs weight normalization to eliminate the interference of the number of samples on the joint evaluation score, and the normalized weight The calculation formula is: ; This process ensures that the sum of the weights of each sample is 1, which is more conducive to stable and reliable data modeling during joint evaluation.

[0039] Step S4 then performs joint evaluation score modeling, and the calculation formula of the joint evaluation score is: ; Wherein, : joint evaluation score, measuring the risk of overall sample anomaly; If The value is proportional to the risk degree of overall sample anomaly. This joint evaluation method overcomes the deficiency of the prior art that each sample is treated the same and is insufficient to depict the overall risk.

[0040] Step S4 finally outputs the joint evaluation score, and the output value is , which is the normalized aggregation of the sample under the weight, more clearly depicting the risk of overall sample set anomaly.

[0041] The prior art usually takes simple summation or average for each sample to obtain an overall anomaly score, lacks joint modeling of sample weight information, and thus the overall evaluation lacks robustness. The method relies on a spatiotemporal correction image to measure the anomaly score of each sample, and then jointly models the sample weight for normalization, thereby forming a joint evaluation score, which more clearly depicts the overall sample abnormal risk. This joint evaluation method significantly overcomes the lack of weight modeling in the prior art, which leads to insufficient robustness, and provides a more stable and reliable data basis for subsequent anomaly detectors.

[0042] S5, normalizing and quantifying the joint evaluation score to obtain a quantized output score; Step S5 normalizes and quantifies the joint evaluation score to provide a stable and reliable data basis for advertising recommendation effect evaluation.

[0043] In order to eliminate the interference of different sample abnormal risk number domains on the quantization result, the joint evaluation score needs to be normalized. The normalization is realized by using the existing mature min-max normalization method, which includes: The maximum score in the joint evaluation score is the upper limit, and the minimum score is the lower limit.

[0044] Each sample is normalized, i.e., linearly normalized to 0-1 according to the maximum and minimum scores.

[0045] After normalization, each sample is assigned a normalized joint evaluation score, which has stable and intuitive numerical distribution and is suitable as a quantization input.

[0046] Finally, the normalized joint evaluation score is quantized and output by using the existing mature linear mapping technology. This process assigns a quantization score to each sample, and the quantization score number domain is 0-100, where the larger the value, the greater the sample abnormal risk. This quantization score provides an intuitive and reliable data basis for advertising recommendation effect evaluation, which helps to measure the overall level of recommendation abnormal risk and provides a quantitative reference for marketing strategy adjustment.

[0047] In advertising recommendation, abnormal risk refers to the risk of recommendation deviation, invalidity, or even negative effect. This risk mainly manifests in the following situations: The recommended content is seriously inconsistent with the user's interest (leading to a decrease in click rate and failure of conversion), thereby wasting marketing resources; The recommendation has negative feedback (such as being shielded by the user, not interesting, and not suitable), which interferes with the subsequent performance of the recommendation model; The recommendation has a negative effect on the overall marketing goal (such as causing waste of marketing funds and a decrease in consumer satisfaction), thereby reducing marketing effectiveness.

[0048] After quantifying the risk as a numerical value through the above steps S1-S6, the risk size is measured in the advertisement recommendation loop, which helps technicians to optimize the model, allocate resources, and adjust the marketing strategy.

[0049] Embodiment Two: Referring to Figure 5 For the second embodiment of the present application, the embodiment is based on the advertisement recommendation effect evaluation system based on multi-dimensional user portrait described in embodiment one, and the function modules include: User portrait quantification module: attribute quantification processing is performed on the collected multi-dimensional user portrait data, discrete, continuous and category type information is converted into a unified portrait vector representation, and a numerical user feature expression is formed for subsequent calculation. This module realizes the standardization and structuring of user portraits, and is the basis for system input information.

[0050] Space-time correction module: using the time and geographic location information of the sample, the user portrait vector is corrected in space-time weight, the influence degree of portrait features is adjusted, the bias caused by space-time heterogeneity is eliminated, and the applicability and accuracy of portrait features under different time and space conditions are ensured. This module realizes the fusion of multi-dimensional portrait and space-time conditions, and improves the dynamic adaptability of the portrait.

[0051] Sample weight correction module: based on the portrait vector corrected in space-time and the occurrence conditions of samples such as clicks and conversions, the sample weight is calculated and adjusted to reflect the relative importance and influence of the sample in the overall recommendation effect evaluation. This module enhances the recognition ability of abnormal samples and key behaviors through comprehensive weight correction, and improves the robustness and accuracy of the evaluation.

[0052] Joint evaluation score calculation module: combining sample weight and space-time corrected portrait features, as well as related occurrence conditions, multi-factor joint evaluation is performed to output a unified joint evaluation score.

[0053] Normalized quantization output module: the joint evaluation score is normalized to standardize the evaluation results in different samples and scenarios into a unified quantization output score, which is convenient for cross-sample, cross-space-time condition comparison and application. The quantization score output by this module can be used as the basis data for subsequent advertisement recommendation decision or risk prompt, which improves the practicability and popularization value of the system.

[0054] Embodiment Three: The embodiment also provides a computer device suitable for the case of the advertisement recommendation effect evaluation system based on multi-dimensional user portraits, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the advertisement recommendation effect evaluation system based on multi-dimensional user portraits as proposed in the above embodiment.

[0055] The embodiment also provides a storage medium having a computer program stored thereon, and the program is executed by a processor to realize the advertisement recommendation effect evaluation system based on multi-dimensional user portraits as proposed in the above embodiment.

[0056] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to perform wired or wireless communication with external terminals. The wireless communication can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or can be a key, trackball or touchpad arranged on the shell of the computer device, or can be an external keyboard, touchpad or mouse, etc.

[0057] If the functions are realized in the form of software function units and sold or used as independent products, the functions can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method of each embodiment of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0058] The logic and / or steps represented in the flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this specification, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium.

[0059] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer. In some embodiments, the computer-readable medium can be non-transitory.

[0060] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques, which are well known in the art of making integrated circuits, can be used to implement the various techniques and technologies described herein: a discrete logic circuit(s) having logic gates for implementing logic functions upon data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array(s) (PGA), a field programmable gate array (FPGA), etc.

[0061] It should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the present application, although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced, without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of claims of the present application.

Claims

1. An advertising recommendation effect evaluation system based on multi-dimensional user portraits, characterized in that: The functional modules of the system include: user portrait quantization module, spatiotemporal correction module, sample weight correction module, joint evaluation score calculation module and normalized quantization output module; The application of the functional modules of the system includes the following steps: Based on the user portrait quantification module, the attributes of the collected user portraits are quantified to obtain the portrait vector; Use the spatiotemporal correction module to perform spatiotemporal correction on the portrait vector obtained by the user portrait quantization module to obtain a spatiotemporal correction portrait vector; The sample weight correction module performs sample weight correction on the spatiotemporal correction image vector output by the spatiotemporal correction module to obtain the sample weight; Based on the sample weights output by the sample weight correction module, a joint evaluation is performed using the joint evaluation score calculation module to obtain a joint evaluation score; The joint evaluation score output by the joint evaluation score calculation module is input into the normalized quantization output module for normalization and quantization to obtain a quantitative output score, which serves as the basis for measuring the recommendation risk in the advertising recommendation loop. The higher the score, the higher the advertising recommendation risk.

2. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 1 is characterized in that: The input portrait in the user portrait quantization module includes: user basic information, user interest tags, user behavior data and user geographic information. The above portrait information is quantified through One-Hot encoding technology, Word Embedding technology and normalization processing to form a portrait vector .

3. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 1 is characterized in that: The spatiotemporal correction module first performs spatiotemporal weight modeling, then performs spatiotemporal deviation correction, and finally performs post-correction image reconstruction; In spatiotemporal weight modeling, spatiotemporal weights are applied to each portrait component. The spatiotemporal weights are determined by the spatiotemporal factors. The corresponding spatiotemporal weights are modeled as spatiotemporal functions, and the formula is: ； in, : spatiotemporal weight; : Collection time; : collection location; : Sigmoid activation function; : The sensitivity of spatiotemporal weight to time; : Sensitivity of spatiotemporal weights to location; : Bias, ensuring stable weight distribution.

4. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 3 is characterized in that: The spatiotemporal correction module eliminates the offset of the image caused by spatiotemporal conditions when performing spatiotemporal deviation correction. This is achieved by correcting each component. The corresponding model formula is: ； in, : After correction A portrait component; : No. before correction A portrait component; : spatiotemporal weights, multiplicative correction for each component.

5. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 3 is characterized in that: When the spatiotemporal correction module reconstructs the corrected image, the corresponding reconstruction formula is: ； in, : The new image vector formed after spatiotemporal correction.

6. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 1, characterized in that: In the sample weight correction module, sample weight modeling is first performed, then sample weight correction is performed, and finally weights are output; When performing sample weight modeling, the sample The anomaly score is: ； in, : No. Sample spatiotemporal correction portrait vector; : The equilibrium vector obtained by the overall mean of the spatiotemporal correction portrait, , where is the total number of samples.

7. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 6, characterized in that: When performing sample weight correction, the sample weight correction module assigns a weight to each sample according to the anomaly score: ； in, : sample weight; :sample Anomaly score; : Weight decay factor.

8. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 6 is characterized in that: The sample weight correction module finally outputs the weight of each sample: ； And thus form the sample weight: 。 9. The advertising recommendation effect evaluation system based on multi-dimensional user portraits according to claim 1, characterized in that: The joint evaluation score calculation module first performs weight normalization, and then performs joint evaluation score modeling to obtain the joint evaluation score .

Images