Method and device for enhancing the interpretability of business results

Abstract

The embodiments of this specification provide a method and device for enhancing the interpretability of business results. The method includes: obtaining an original behavior sequence and a corresponding business label, wherein the original behavior sequence includes T operation behaviors arranged in chronological order by a target user; using a trained business prediction model, determining M business prediction results corresponding to M first sequences; the M first sequences are obtained by respectively shielding M groups of behaviors from the original behavior sequence; based on the business label, determining M prediction losses corresponding to the M business prediction results; based on the M prediction losses, determining key behaviors from the M groups of behaviors, which are used to explain the business prediction results of the original behavior sequence.

Description

Technical Field

[0001] One or more embodiments of the present specification relate to the field of machine learning technology, and more particularly, to a method and apparatus for enhancing the interpretability of business results. Background Art

[0002] In many business scenarios, it is necessary to analyze and process user operation behaviors, for example, to assess the risk level associated with the user in order to conduct risk prevention and control. Usually, analysis can be performed based on the characteristics of a certain operation performed by the user. Furthermore, the user's behavior sequence can be considered more comprehensively. The behavior sequence is the process of a series of clicks, visits, purchases, etc. generated by the user in daily operation. It can be expressed as a time series of behavior sets. It contains the user's fine-grained habit preferences and other characteristics, which facilitates a more comprehensive analysis of the user's operation history and operation mode.

[0003] With the development of machine learning, building machine learning models to analyze user behavior sequences has become a research hotspot. In some cases, it is necessary to explain the business prediction results for user behavior sequences obtained using machine learning models. For example, in risk prevention and control scenarios, when users ask customer service staff about the reasons because they feel that their operation behavior has been interfered with or blocked, customer service staff need to explain to the users.

[0004] However, the current interpretation of business analysis results mostly relies on manual analysis of user behavior sequences, which results in high costs and very limited efficiency and accuracy. Therefore, there is an urgent need for a solution that can effectively enhance the interpretability of business results. Summary of the invention

[0005] One or more embodiments of this specification describe a method for enhancing the interpretability of business results, which can automatically determine key behaviors in an original behavior sequence and / or key attributes of the behaviors, thereby effectively enhancing the interpretability of business results.

[0006] According to a first aspect, a method for enhancing the interpretability of business results is provided, comprising: obtaining an original behavior sequence and a corresponding business label, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order; using a trained business prediction model, determining M business prediction results corresponding to M first sequences; the M first sequences are obtained by respectively shielding M groups of behaviors from the original behavior sequence; based on the business label, determining M prediction losses corresponding to the M business prediction results; based on the M prediction losses, determining key behaviors from the M groups of behaviors, for explaining the business prediction results of the original behavior sequence.

[0007] In one embodiment, the service tag is obtained through manual annotation.

[0008] In one embodiment, obtaining the original behavior sequence and the corresponding business label includes: obtaining the original behavior sequence; and using the business prediction model to determine the business prediction result of the original behavior sequence as the business label.

[0009] In one embodiment, the business tag includes a plurality of sub-tags corresponding to a plurality of prediction tasks, and the business prediction result includes a plurality of sub-results corresponding to the plurality of prediction tasks.

[0010] In one embodiment, the business prediction model includes a behavioral attention layer; wherein, before using the trained business prediction model to determine M business prediction results corresponding to M first sequences, the method also includes: obtaining T attention weights corresponding to the T operation behaviors determined using the behavioral attention layer; determining some of the T attention weights that are greater than a weight threshold or are within a preset front range; and grouping the operation behaviors corresponding to the some attention weights to obtain the M groups of behaviors.

[0011] In one embodiment, any two groups of behaviors in the M groups of behaviors are mutually exclusive.

[0012] In one embodiment, before using the trained business prediction model to determine M business prediction results corresponding to M first sequences, the method also includes: for any group of behaviors in the M groups of behaviors, replacing the group of behaviors included in the original behavior sequence with default values ​​to obtain the corresponding first sequence.

[0013] In one embodiment, the business prediction model includes a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, M business prediction results corresponding to M first sequences are determined, including: using the behavior representation layer to determine T behavior representation vectors corresponding to the T operation behaviors; for any group of behaviors in the M groups of behaviors, setting the behavior representation vectors corresponding to the group of behaviors contained in the T behavior representation vectors to zero; using the sequence representation layer, processing the T behavior representation vectors obtained after the zeroing process to obtain the sequence representation vector of the first sequence corresponding to the group of behaviors; using the output layer to process the sequence representation vector to obtain the business prediction result corresponding to the first sequence.

[0014] In one embodiment, the business prediction model includes a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, M business prediction results corresponding to M first sequences are determined, including: using the behavior representation layer to determine T behavior representation vectors corresponding to the T operation behaviors; for any group of behaviors among the M groups of behaviors, setting the model parameters in the sequence representation layer used to process the group of behaviors to zero; using the sequence representation layer after the zeroing process to process the T behavior representation vectors to obtain the sequence representation vector of the first sequence corresponding to the group of behaviors; using the output layer to process the sequence representation vector to obtain the business prediction result corresponding to the first sequence.

[0015] In one embodiment, based on the M predicted losses, key behaviors are determined from the M groups of behaviors, including: determining several predicted losses among the M predicted losses that are greater than a loss threshold or ranked within a preset front range; and determining several groups of behaviors corresponding to the several predicted losses as the key behaviors.

[0016] In a specific embodiment, each of the T operation behaviors corresponds to N attribute fields; wherein, after determining the key behavior from the M groups of behaviors based on the M predicted losses, the method further includes: using the business prediction model to determine Q business prediction results corresponding to Q second sequences; the Q second sequences correspond to sequences obtained by respectively masking Q groups of attribute fields in the behavior sequence formed by the key behaviors; based on the business label, determining Q predicted losses corresponding to the Q business prediction results; based on the Q predicted losses, determining the key attribute field from the Q groups of attribute fields, which is used to explain the business prediction results of the original behavior sequence.

[0017] According to the second aspect, a method for enhancing the interpretability of business results is provided, comprising: obtaining an original behavior sequence and a corresponding business label, the original behavior sequence comprising T operation behaviors performed by a target user and arranged in chronological order, wherein each operation behavior comprises N attribute fields; using a trained business prediction model, determining S business prediction results corresponding to S third sequences; the S third sequences are obtained by respectively masking S groups of attribute fields from the original behavior sequence; based on the business label, determining S prediction losses corresponding to the S business prediction results; based on the S prediction losses, determining key attribute fields from the S groups of attribute fields for explaining the business prediction results of the original behavior sequence.

[0018] In one embodiment, any two groups of attribute fields in the S groups of attribute fields are mutually exclusive.

[0019] In one embodiment, before using the trained business prediction model to determine S business prediction results corresponding to S third sequences, the method also includes: for any group of attribute fields in the S groups of attribute fields, replacing the field value of the attribute field in the group corresponding to each behavior in the original behavior sequence with a default value to obtain the corresponding third sequence.

[0020] In one embodiment, the business prediction model includes an attribute coding layer, a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, S business prediction results corresponding to S third sequences are determined, including: for each operation behavior among the T operation behaviors, using the attribute coding layer to process the field values ​​of the corresponding N attribute fields contained in the behavior to obtain the corresponding N attribute coding vectors; for any group of attribute fields among the S groups of attribute fields, setting the attribute coding vectors corresponding to the group of attribute fields in the N attribute coding vectors to zero; using the behavior representation layer, processing the N attribute coding vectors obtained after the zeroing process to obtain the behavior representation vector corresponding to the behavior; using the sequence representation layer, processing the T behavior representation vectors corresponding to the T operation behaviors to obtain the sequence representation vector of the third sequence corresponding to the group of attribute fields; using the output layer to process the sequence representation vector to obtain the business prediction result corresponding to the third sequence.

[0021] In one embodiment, the business prediction model includes an attribute coding layer, a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, S business prediction results corresponding to S third sequences are determined, including: for each operation behavior among the T operation behaviors, using the attribute coding layer to process the field values ​​of the corresponding N attribute fields contained in the behavior to obtain corresponding N attribute coding vectors; for any group of attribute fields among the S groups of attribute fields, setting the model parameters used for processing the group of attribute fields in the behavior representation layer to zero; using the behavior representation layer after the zeroing process, processing the N attribute coding vectors to obtain the behavior representation vector corresponding to the behavior; using the sequence representation layer to process the T behavior representation vectors corresponding to the T operation behaviors to obtain the sequence representation vector of the third sequence corresponding to the group of attribute fields; using the output layer to process the sequence representation vector to obtain the business prediction result corresponding to the third sequence.

[0022] In one embodiment, based on the S predicted losses, key attribute fields are determined from the S groups of attribute fields, including: determining several predicted losses among the S predicted losses that are greater than a loss threshold or ranked within a preset front range; and determining several groups of attribute fields corresponding to the several predicted losses as the key attribute fields.

[0023] According to a third aspect, a device for enhancing the interpretability of business results is provided, comprising: a sequence acquisition unit, configured to acquire an original behavior sequence and a corresponding business label, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order; a result prediction unit, configured to determine M business prediction results corresponding to M first sequences using a trained business prediction model; the M first sequences are obtained by respectively shielding M groups of behaviors from the original behavior sequence; a loss determination unit, configured to determine M predicted losses corresponding to the M business prediction results based on the business label; and a behavior determination unit, configured to determine key behaviors from the M groups of behaviors based on the M predicted losses, for explaining the business prediction results of the original behavior sequence.

[0024] According to a fourth aspect, a device for enhancing the interpretability of business results is provided, comprising: a sequence acquisition unit, configured to acquire an original behavior sequence and a corresponding business label, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order, wherein each operation behavior includes N attribute fields; a result prediction unit, configured to use a trained business prediction model to determine S business prediction results corresponding to S third sequences; the S third sequences are obtained by respectively masking S groups of attribute fields from the original behavior sequence; a loss determination unit, configured to determine S predicted losses corresponding to the S business prediction results based on the business label; and an attribute determination unit, configured to determine a key attribute field from the S groups of attribute fields based on the S predicted losses, for explaining the business prediction results of the original behavior sequence.

[0025] According to a fifth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed in a computer, the computer is caused to execute the method of the first aspect or the second aspect.

[0026] According to a sixth aspect, a computing device is provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method of the first aspect or the second aspect is implemented.

[0027] By using the method and device provided in the embodiments of this specification, the impact of each group of behaviors included in the original behavior sequence on the business prediction results is evaluated by calculating the prediction loss, so as to determine the key behaviors and / or key losses that support the business prediction results of the original behavior sequence, and effectively enhance the interpretability of the business prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 A schematic diagram showing an implementation architecture of a solution for enhancing the explainability of business results disclosed in an embodiment of this specification;

[0030] Figure 2 A schematic diagram showing a method flow for enhancing the explainability of business results according to an embodiment;

[0031] Figure 3 A schematic diagram showing a model structure of a business prediction model according to an embodiment;

[0032] Figure 4 A schematic diagram showing a model structure of a business prediction model according to another embodiment;

[0033] Figure 5 A schematic flow chart showing a method for enhancing the explainability of business results according to another embodiment;

[0034] Figure 6 A schematic diagram showing a model structure of a business prediction model according to yet another embodiment;

[0035] Figure 7 A schematic diagram showing a model structure of a business prediction model according to another embodiment;

[0036] Figure 8 A schematic diagram showing the structure of an apparatus for enhancing the interpretability of business results according to an embodiment;

[0037] Fig. 9 A schematic diagram of the structure of an apparatus for enhancing the explainability of business results according to another embodiment is shown. DETAILED DESCRIPTION

[0038] The solution provided in this specification is described below in conjunction with the accompanying drawings.

[0039] As mentioned earlier, there is a need to enhance the explainability of business results. Figure 1 FIG. 1 is a schematic diagram showing an implementation architecture of a solution for enhancing the interpretability of business results disclosed in an embodiment of this specification. Figure 1 As shown in FIG. 1 , during the implementation of the solution, the original behavior sequence and the corresponding business tag are first obtained, wherein the original behavior sequence includes T operation behaviors (or operation events) arranged in chronological order, namely, E1 to E T; Then, the trained business prediction model is used to determine the M prediction results corresponding to the M first sequences obtained by respectively shielding the M groups of behaviors from the original behavior sequence, Figure 1 The blocked behaviors are shown by the diagonal shaded box; then, the key behaviors are determined from the above M groups of behaviors using the M prediction losses determined based on the business labels and the M prediction results. Figure 1 In the figure, it is shown as E1, E3, E5 and E T In this way, the key behaviors in the original behavior sequence can be automatically determined to explain the business prediction results (or simply business results, prediction results) of the original behavior sequence, thereby effectively enhancing the interpretability of the business results.

[0040] The implementation steps of the above scheme are described in detail below. Figure 2 A schematic diagram of a method flow for enhancing the interpretability of business results according to an embodiment is shown. The execution subject of the method can be any device, server or device cluster with computing and processing capabilities. Figure 2 As shown, the method comprises the following steps:

[0041] Step S210, obtaining an original behavior sequence and a corresponding business label, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order; Step S220, using a trained business prediction model, determining M business prediction results corresponding to M first sequences; the M first sequences are obtained by respectively shielding M groups of behaviors from the original behavior sequence; Step S230, based on the business label, determining M prediction losses corresponding to the M business prediction results; Step S240, based on the M prediction losses, determining key behaviors from the M groups of behaviors, which are used to explain the business prediction results of the original behavior sequence.

[0042] The above steps are expanded as follows:

[0043] First, in step S210, the original behavior sequence and the corresponding service label are obtained.

[0044] See also Figure 1 The original behavior sequence includes T operation behaviors E1, E2, …, E1 arranged in chronological order. T; and, these T operation behaviors are all performed by the target user, where T is an integer greater than 1. In one embodiment, the T operation behaviors are all historical operation behaviors of the target user; in another embodiment, the T operation behaviors may include the current operation behavior of the target user and T-1 historical operation behaviors traced back. On the other hand, in one embodiment, T operation behaviors of the same type may be collected, such as login-type behaviors or transaction-type behaviors; in another embodiment, T operation behaviors may be collected according to a plurality of pre-set behavior types, such as click behaviors, purchase behaviors, etc.

[0045] Each operation behavior in the original behavior sequence has related attribute information, including N attribute values ​​corresponding to N attribute fields, where N is an integer greater than 1. For the comprehensiveness of business analysis, fine-grained and comprehensive attribute information of each behavior can be obtained here for subsequent processing. Exemplarily, the attribute information may include the behavior type (such as login operation, recharge operation, payment operation, transaction operation, etc.), behavior time, geographic location, device information used by the user operation (such as device type, device model, MAC address, IP address, etc.), and information about the software used (such as browser type, app version, etc.); further, if the operation behavior is a transaction behavior, the attribute information may also include the amount involved in the transaction behavior, payment channel, payment card number, payment card issuing bank, transaction object, etc.

[0046] In addition, for historical operation events, the corresponding attribute information may also include the operation results of historical operation behaviors, such as operation success, failure, timeout, etc. For current operation events, they are generally events to be analyzed and predicted, and they have not yet generated operation results. The field value of the corresponding operation result may be a default value; generally speaking, business analysis of the current operation event is used to determine whether to allow this operation, that is, to determine its operation result.

[0047] The original behavior sequence is introduced above. The business tag corresponding to the original behavior sequence is related to the business prediction task to be performed. The business object targeted by the business prediction task can be a user or the user's current operation behavior. In one embodiment, the business prediction task is a risk assessment task for the user or the current operation, and the business tag is a risk category tag, such as high risk, medium risk, low risk, or risky and risk-free, etc.; in another embodiment, the business prediction task is a credit prediction task for the user, and the business tag can be a credit score or credit rating. On the other hand, the business prediction task to be performed can also include multiple items, and the business tags corresponding to the original behavior sequence are multiple. In one embodiment, the business prediction task includes identifying risky users and selecting behavioral intervention methods for high-risk users. Accordingly, the business tags include risk category tags, such as high risk and low risk, and also include intervention category tags, such as telephone inquiries, requiring secondary verification, SMS reminders, etc.

[0048] On the other hand, regarding the acquisition method of the above-mentioned service tag, in one embodiment, the service tag obtained by manual annotation can be directly acquired. In another embodiment, in the case where there is no manually labeled service tag, for example, the above-mentioned original behavior sequence responds to the current operation behavior E T However, there is no time for manual labeling during data collection. In this case, the original behavior sequence can be input into the trained business prediction model to obtain the corresponding business prediction result as the business label corresponding to the original behavior sequence.

[0049] Based on this, in step S220, the trained business prediction model is used to determine M business prediction results corresponding to M first sequences, where the M first sequences are obtained by respectively shielding M groups of behaviors from the original business sequence.

[0050] It should be noted that the above business prediction model can be obtained by training in a supervised learning manner. Specifically, multiple historical behavior sequences and corresponding multiple manually annotated labels can be collected, and then, based on the collected training data, the business prediction model can be trained for multiple rounds to obtain a trained business prediction model.

[0051] For ease of understanding, the following first introduces the method for determining the above-mentioned M groups of blocked behaviors. It should be noted that in one implementation scenario, the business object targeted by the business prediction task is the current operation behavior E T At this time, we can remove E from the above T operations. TIn another implementation scenario, the business prediction task targets the user. In this case, group processing can be performed based on the T operation behaviors. For the sake of simplicity, the following mainly uses group processing based on T operation behaviors as an example. The case of group processing based on T-1 operation behaviors can be executed by reference, including replacing T with T-1.

[0052] In one embodiment, T operation behaviors can be directly divided into M groups, each group can include one or more operation behaviors, and the number of operation behaviors in the group is less than T, or less than a predetermined number threshold, where the number threshold is usually a value less than T. Further, in a specific embodiment, the T operation behaviors can be divided into equal amounts so that each group contains the same number of behaviors. In one example, each of the T operation behaviors can be separately regarded as a group of behaviors, in which case M=T. On the other hand, in a specific embodiment, any two groups of behaviors are mutually exclusive, that is, any two groups of behaviors do not contain the same operation behaviors.

[0053] In another embodiment, the T operation behaviors may be pre-screened first, and then the M groups of behaviors may be determined based on the operation behaviors retained after the pre-screening. Figure 3 As shown, the business prediction model includes a behavior attention layer 32, based on which T attention weights corresponding to the above-mentioned T operation behaviors determined by using the behavior attention layer 32 can be obtained; then, some attention weights among the T attention weights that are greater than the weight threshold or are within a preset front range are determined; thereafter, the operation behaviors corresponding to these some attention weights are grouped to obtain the above-mentioned M groups of behaviors.

[0054] Further, in a specific embodiment, see Figure 3 , the acquisition of T attention weights may include: using the behavior representation layer 31 to respectively represent the T operation behaviors E1, E2, ..., E T Processing is performed to obtain the corresponding T behavior representation vectors X1, X2, …, X T ; Then, in the behavior attention layer 32, the T behavior representation vectors are processed to obtain the corresponding T attention weights w1, w2, ..., w T ( Figure 3 The T attention weights are used to perform weighted summation on the T behavior representation vectors to obtain the behavior sequence representation vector V. It should be understood that the determination of the attention weights in the behavior attention layer 32 can be implemented using an existing attention mechanism, which will not be described in detail here. Thus, T attention weights can be obtained. In addition, it should be noted that for the case of pre-screening based on the above T-1 operation behaviors, the attention weights can be based on Figure 4The behavior shown in the attention layer 42 is determined.

[0055] In a specific embodiment, the weight threshold may be manually set. In a specific embodiment, the preset top range may be a predetermined number of tops, such as top 5, or a predetermined percentage of tops, such as top 50%.

[0056] In a specific embodiment, the operation behaviors corresponding to the above-mentioned partial attention weights are grouped, such as randomly divided or equally divided, and reference may be made to the above-mentioned related description for this.

[0057] Thus, M groups of behaviors can be determined. Further, the business prediction results of the M first sequences corresponding to the M groups of behaviors can be determined. It should be noted that the masking of the information contained in each group of behaviors can be achieved from multiple dimensions, for example, the behavior data obtained by the input layer, the characterization vector corresponding to the behavior data, or the model parameters used to process the characterization vector are masked. Specifically, in one embodiment, for any group of behaviors in the M groups of behaviors, the group of behaviors contained in the above original behavior sequence is replaced with a default value (such as null) to obtain the corresponding first sequence; then, the obtained M first sequences corresponding to the M groups of behaviors can be respectively input into the trained business prediction model to obtain the corresponding M business prediction results. In this way, the business prediction results of the corresponding partially masked sequences of behaviors can be determined by masking the original information of each group of behaviors.

[0058] In another embodiment, the business prediction model includes a behavior representation layer, a sequence representation layer, and an output layer. It should be understood that each layer in the business prediction model can be implemented using a neural network. In one example, the sequence representation layer can be implemented as Figure 3 The behavior attention layer 32 shown in FIG. 3 , correspondingly, the behavior representation layer and the output layer can be implemented as Figure 3 The behavior representation layer 31 and the output layer 33 are shown; in another example, the sequence representation layer can be implemented as Figure 4 The behavior attention layer 42 shown in FIG. 4 , correspondingly, the behavior representation layer and the output layer can be implemented as Figure 4 A behavior representation layer 41 and an output layer 43 are shown.

[0059] Furthermore, in a specific embodiment, the determination of the M business prediction results may include: firstly using the behavior representation layer to determine T behavior representation vectors X1, X2, ..., X corresponding to the T operation behaviors. T Then, for any group of the above M groups of behaviors, the T behavior representation vectors X1, X2, ..., X TThe behavior representation vectors corresponding to the group of behaviors are set to zero. For example, assuming that a group of behaviors includes operation behavior E1, its corresponding behavior representation vector X1 can be replaced by a zero vector, thereby obtaining T behavior representation vectors 0, X2, …, X1 that are set to zero. T ; Then, the sequence representation layer is used to process the T behavior representation vectors obtained after the zeroing process to obtain the sequence representation vector of the first sequence corresponding to the group of behaviors; then, the output layer is used to process the sequence representation vector to obtain the business prediction result corresponding to the first sequence. In this way, the business prediction result of the corresponding partially masked sequence of behaviors can be determined by masking the representation vectors corresponding to each group of behaviors.

[0060] In another specific embodiment, the determination of the above-mentioned M business prediction results may include: first using the behavior representation layer to determine the T behavior representation vectors corresponding to the above-mentioned T operation behaviors; further, for any group of behaviors in the M groups of behaviors, the model parameters used to process the group of behaviors in the sequence representation layer are set to zero, and the sequence representation layer after the zeroing process is used to process the T behavior representation vectors to obtain the sequence representation vector of the first sequence corresponding to the group of behaviors, and then the output layer is used to process the sequence representation vector to obtain the business prediction result corresponding to the first sequence. In this way, the business prediction result of the corresponding partially blocked sequence of behaviors can be determined by shielding the model parameters used to process the representation vectors of each group of behaviors.

[0061] From the above, M business prediction results corresponding to the M first sequences can be determined. Then, in step S230, based on the business labels corresponding to the above original behavior sequences, M predicted losses corresponding to the M business prediction results are determined. Specifically, for any of the business prediction results, it and the business label are input into a preset loss function, and the output result of the loss function is used as the predicted loss corresponding to the business prediction result. In this way, M predicted losses corresponding to the M business results can be obtained.

[0062] Then, in step S240, based on the above-mentioned M predicted losses, key behaviors are determined from the M groups of behaviors. In one embodiment, several predicted losses greater than the loss threshold among the M predicted losses can be determined, and then several groups of behaviors corresponding to the several predicted losses are determined as the above-mentioned key behaviors. It should be understood that the loss threshold can be manually preset. In another embodiment, the M predicted losses can be sorted from large to small, so that several groups of behaviors corresponding to several predicted losses ranked in a preset front range are determined as the above-mentioned key behaviors. It should be understood that the preset front range can be manually preset, and specifically can be a ranking range, such as top5, or can also be a percentage range, such as top30%.

[0063] In this way, by using prediction loss as an evaluation indicator of behavior importance, the key behaviors that support the business prediction results of the original behavior prediction can be determined to explain the business prediction results of the original behavior sequence.

[0064] It should be noted that the above mainly introduces the scenario where the business prediction model is used to perform a single prediction task. In fact, the solution disclosed in the embodiments of this specification is also applicable to the scenario where the business prediction model is used to perform multiple prediction tasks. In this scenario, the above business tag includes multiple sub-tags corresponding to the multiple prediction tasks, and the above business prediction results include multiple sub-results corresponding to the multiple prediction tasks. Further, in one embodiment, the model structure of the business prediction model may include Figure 3 The multiple output layers 33 shown in FIG. 1 , or Figure 4 The multiple output layers 43 shown in the figure are used to output the prediction results of the corresponding prediction tasks respectively.

[0065] On the other hand, in one embodiment, when determining the prediction loss corresponding to each business prediction result, the sub-loss corresponding to each prediction task can be determined first, and then the comprehensive loss corresponding to multiple prediction tasks can be determined as the prediction loss corresponding to the business prediction result. Specifically, first determine the corresponding sub-loss based on the sub-label corresponding to each prediction task and the sub-result corresponding to the prediction task in the business prediction result; then perform weighted summation on the multiple sub-losses corresponding to the multiple prediction tasks to obtain the corresponding prediction loss. In one example, the weights used in the weighted summation can be manually pre-set, for example, setting the same weights for multiple prediction tasks. In another example, the weights of the weighted summation can be training parameters, which are obtained by training together with the business prediction model.

[0066] The above is an introduction to the application of the solution disclosed in this specification to multi-task prediction scenarios.

[0067] In summary, the method for enhancing the interpretability of business results disclosed in the embodiments of this specification is adopted to evaluate the impact of each group of behaviors included in the original behavior sequence on the business prediction results by calculating the prediction loss, thereby determining the key behaviors that support the business prediction results of the original behavior sequence, and effectively enhancing the interpretability of the business prediction results.

[0068] The above mainly introduces the method for determining the key behavior in the original behavior sequence. According to another embodiment, the embodiment of this specification also discloses a method for determining the key attributes in the behavior, and the determined key attributes can also be used to explain the business prediction results of the original behavior sequence.

[0069] Figure 5A schematic flow chart of a method for enhancing the interpretability of business results according to another embodiment is shown. The execution subject of the method can be implemented as any server, device or equipment cluster with computing and processing capabilities. Figure 5 As shown, the method comprises the following steps:

[0070] Step S510, obtaining an original behavior sequence and a corresponding business label, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order, wherein each operation behavior includes N attribute fields; Step S520, using a trained business prediction model, determining S business prediction results corresponding to S third sequences, wherein the S third sequences are obtained by respectively masking S groups of attribute fields from the original behavior sequence; Step S530, based on the business label, determining S prediction losses corresponding to the S business prediction results; Step S540, based on the S prediction losses, determining a key attribute field from the S groups of attribute fields, which is used to explain the business prediction results of the original behavior sequence.

[0071] The above steps are expanded as follows:

[0072] First, in step S510, an original behavior sequence and a corresponding service tag are obtained, wherein the original behavior sequence includes T operation behaviors made by the target user arranged in chronological order, wherein each operation behavior includes N attribute fields. It should be noted that the introduction to step S510 can refer to the introduction to step S210 above, and will not be repeated here.

[0073] Next, in step S520, the trained business prediction model is used to determine S business prediction results corresponding to S third sequences, wherein the S third sequences are obtained by respectively masking S groups of attribute fields from the original behavior sequence.

[0074] It should be noted that the above business prediction model can be obtained by training using a supervised learning method.

[0075] For ease of understanding, the following first introduces the method for determining S groups of attribute fields. In one embodiment, the above-mentioned N attribute fields can be directly divided into S groups, each group can include one or more attribute fields, and the number of attribute fields in the group is less than N, or less than a predetermined number threshold, where the number threshold is usually a value less than N. Further, in a specific embodiment, the N attribute fields can be divided equally so that each group contains the same number of attribute fields. In one example, each attribute field in the N attribute fields can be separately regarded as a group of attribute fields, in which case S=N. On the other hand, in a specific embodiment, any two groups of attribute fields are mutually exclusive, that is, there are no identical attribute fields between any two groups of attribute fields. In this way, S groups of attribute fields can be determined.

[0076] Further, the business prediction results of the S first sequences corresponding to the above-mentioned S groups of attribute fields can be determined. It should be noted that the information contained in each group of attribute fields can be shielded from multiple dimensions, for example, the attribute field value obtained by the input layer, or the encoding vector corresponding to the attribute field value, or the model parameter used to process the encoding vector is shielded. Specifically, in one embodiment, for any one of the S groups of attribute fields, the field value of the attribute field of each behavior in the above-mentioned original behavior sequence corresponding to the group is replaced with a default value (such as null) to obtain a corresponding third sequence; then, the obtained S third sequences corresponding to the S groups of attribute fields can be respectively input into the trained business prediction model to obtain the corresponding S business prediction results. In this way, the business prediction results of the corresponding partially attribute-shielded sequence can be determined by shielding the field values ​​of each group of attribute fields.

[0077] In another embodiment, the business prediction model includes an attribute encoding layer, a behavior representation layer, a sequence representation layer, and an output layer. In one example, the business prediction model may include Figure 6 The attribute encoding layer 61, the behavior representation layer 62, the sequence representation layer 63 and the output layer 64 shown in FIG. Figure 3 The behavior representation layer 31 shown in FIG. 3 can be used to implement Figure 6 In another example, the business prediction model may include: Figure 7 The attribute encoding layer 71, the behavior representation layer 72, the sequence representation layer 73 and the output layer 74 shown in FIG. Figure 4 The behavior representation layer 41 shown in FIG. 4 can be used to implement Figure 7 The functions of the attribute encoding layer 71 and the behavior representation layer 72.

[0078] Based on this, in a specific embodiment, the determination of the S service prediction results may include: for each behavior E in the T behaviors,i , using the attribute encoding layer to process the N field values ​​F1, F2, ..., F corresponding to the N attribute fields contained in the behavior N , and obtain the corresponding N attribute encoding vectors f1,f2,…,f N ; Then, for any one of the S groups of attribute fields, the attribute coding vectors corresponding to the group of attribute fields in the N attribute coding vectors are set to zero. For example, assuming that a group of attribute fields includes attribute field F2, the corresponding attribute coding vector f2 can be replaced with a zero vector, thereby obtaining N attribute coding vectors f1,0,…,f1 that are set to zero. N ; Then, the behavior representation layer is used to process the N attribute encoding vectors obtained after zeroing to obtain the behavior E i The corresponding behavior representation vector X i ; Then, using the sequence representation layer, the T behavior representation vectors X1, X2, …, X corresponding to the above T operation behaviors are T Processing is performed to obtain a sequence representation vector V of the third sequence corresponding to the group of attribute fields; then, the sequence representation vector V is processed by the output layer to obtain a business prediction result corresponding to the third sequence. In this way, the business prediction result of the corresponding partially attribute-shielded sequence can be determined by masking the attribute encoding vectors corresponding to each group of attribute fields.

[0079] In another specific embodiment, the determination of the S service prediction results may include: for each operation behavior E in the T operation behaviors, i , using the attribute coding layer to process the N field values ​​F1, F2, ..., F corresponding to the N attribute fields contained in the operation behavior N , and obtain the corresponding N attribute encoding vectors f1,f2,…,f N ; Then, for any group of attribute fields in the S group of attribute fields, the model parameters used to process the group of attribute fields in the behavior representation layer are set to zero; then, the N attribute encoding vectors are processed using the behavior representation layer after the zeroing process to obtain the behavior E i The corresponding behavior representation vector X i ; Then, using the sequence representation layer, the T behavior representation vectors X1, X2, …, X corresponding to the T operation behaviors are T Processing is performed to obtain a sequence representation vector V of the third sequence corresponding to the group of attribute fields; then, the sequence representation vector V is processed by the output layer to obtain a business prediction result corresponding to the third sequence. In this way, the business prediction result of the corresponding partially blocked sequence can be determined by masking the model parameters used to process the encoding vectors corresponding to each group of attribute fields.

[0080] From the above, S business prediction results corresponding to the S third sequences can be determined. Then, in step S530, based on the business tags, the S prediction losses corresponding to the S business prediction results are determined. Thus, in step S540, based on the S prediction losses, key attribute fields are determined from the S groups of attribute fields to explain the business prediction results of the original behavior sequence. It should be noted that the description of steps S530 and S540 can refer to the aforementioned introduction to steps S230 and S240, which will not be repeated here.

[0081] In summary, the method for enhancing the interpretability of business results disclosed in the embodiments of this specification is adopted to evaluate the impact of each group of behavior attributes involved in the original behavior sequence on the business prediction results by calculating the prediction loss, thereby determining the key attributes of the business prediction results supporting the original behavior sequence, and effectively enhancing the interpretability of the business prediction results.

[0082] In the above embodiments, the scheme for determining key behaviors and the scheme for key attributes are described separately to enhance the interpretability of business results. In fact, the two schemes can also be used in a superimposed manner. According to an embodiment of one aspect, the key behaviors can be determined first, and then the key attributes can be determined based on the behavior sequence formed by the determined key behaviors, thereby determining the key behaviors and key attributes.

[0083] Specifically, in Figure 2 After step S240 shown, the method may further include: using the above-mentioned business prediction model to determine Q business prediction results corresponding to Q second sequences, wherein the Q second sequences correspond to sequences obtained by respectively shielding Q groups of attribute fields in the behavior sequence formed by the above-mentioned key behaviors; then, based on the above-mentioned business tags, determining Q prediction losses corresponding to the Q business prediction results; then, based on the Q prediction losses, determining key attribute fields from the Q groups of attribute fields for explaining the business prediction results of the original behavior sequence. It should be noted that a detailed introduction to this embodiment can be found in Figure 5 The description of the method is shown and will not be repeated here.

[0084] According to another embodiment, the key attribute may be determined first, and then the key attribute may be determined based on a behavior sequence formed by behaviors including the key attribute, thereby determining the key attribute and the key behavior.

[0085] From the above, the key behaviors and key attributes in the original behavior sequence can be determined in turn, thereby further enhancing the interpretability of the business prediction results of the original behavior sequence.

[0086] Corresponding to the above-mentioned method for enhancing explanatory power, the embodiments of this specification also disclose an apparatus for enhancing explanatory power.

[0087] Figure 8A schematic diagram of a device structure for enhancing the interpretability of business results according to an embodiment is shown. Figure 8 As shown, the device 800 includes the following units:

[0088] The sequence acquisition unit 810 is configured to acquire an original behavior sequence and a corresponding business tag, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order; the result prediction unit 820 is configured to use a trained business prediction model to determine M business prediction results corresponding to M first sequences; the M first sequences are obtained by respectively shielding M groups of behaviors from the original behavior sequence; the loss determination unit 830 is configured to determine M predicted losses corresponding to the M business prediction results based on the business tags; the behavior determination unit 840 is configured to determine key behaviors from the M groups of behaviors based on the M predicted losses, so as to explain the business prediction results of the original behavior sequence.

[0089] In one embodiment, the service tag is obtained through manual annotation.

[0090] In one embodiment, the sequence acquisition unit 810 is configured to: acquire the original behavior sequence; and use the business prediction model to determine the business prediction result of the original behavior sequence as the business label.

[0091] In one embodiment, the business tag includes a plurality of sub-tags corresponding to a plurality of prediction tasks, and the business prediction result includes a plurality of sub-results corresponding to the plurality of prediction tasks.

[0092] In one embodiment, the business prediction model includes a behavioral attention layer; the device 800 also includes a behavior grouping unit 850, which is configured to: obtain T attention weights corresponding to the T operation behaviors determined using the behavioral attention layer; determine some of the T attention weights that are greater than a weight threshold or are within a preset front range; group the operation behaviors corresponding to the some attention weights to obtain the M groups of behaviors.

[0093] In one embodiment, any two groups of behaviors in the M groups of behaviors are mutually exclusive.

[0094] In one embodiment, the apparatus 800 further includes: a sequence generating unit 860 configured to replace any group of behaviors among the M groups of behaviors with default values ​​in the original behavior sequence to obtain a corresponding first sequence.

[0095] In one embodiment, the result prediction unit 820 is specifically configured as follows: using the behavior representation layer to determine the T behavior representation vectors corresponding to the T operation behaviors; for any group of behaviors among the M groups of behaviors, setting the behavior representation vectors corresponding to the group of behaviors contained in the T behavior representation vectors to zero; using the sequence representation layer to process the T behavior representation vectors obtained after the zeroing process to obtain a sequence representation vector of the first sequence corresponding to the group of behaviors; using the output layer to process the sequence representation vector to obtain a business prediction result corresponding to the first sequence.

[0096] In one embodiment, the business prediction model includes a behavior representation layer, a sequence representation layer and an output layer; the result prediction unit 820 is specifically configured as follows: using the behavior representation layer, determining T behavior representation vectors corresponding to the T operation behaviors; for any group of behaviors among the M groups of behaviors, setting the model parameters in the sequence representation layer used to process the group of behaviors to zero; using the sequence representation layer after the zeroing process to process the T behavior representation vectors to obtain the sequence representation vector of the first sequence corresponding to the group of behaviors; using the output layer to process the sequence representation vector to obtain the business prediction result corresponding to the first sequence.

[0097] In one embodiment, the behavior determination unit 840 is specifically configured to: determine several predicted losses among the M predicted losses that are greater than a loss threshold or ranked within a preset front range; and determine several groups of behaviors corresponding to the several predicted losses as the key behaviors.

[0098] In a specific embodiment, each of the T operation behaviors corresponds to N attribute fields; the device 800 also includes an attribute determination unit 870, configured to: use the business prediction model to determine Q business prediction results corresponding to Q second sequences; the Q second sequences correspond to sequences obtained by respectively masking Q groups of attribute fields in the behavior sequence formed by the key behaviors; based on the business label, determine Q prediction losses corresponding to the Q business prediction results; based on the Q prediction losses, determine the key attribute field from the Q groups of attribute fields, which is used to explain the business prediction results of the original behavior sequence.

[0099] Fig. 9 FIG. 2 is a schematic diagram showing a device structure for enhancing the interpretability of business results according to another embodiment. Fig. 9 As shown, the device 900 includes the following units:

[0100] A sequence acquisition unit 910 is configured to acquire an original behavior sequence and a corresponding business tag, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order, wherein each operation behavior includes N attribute fields; a result prediction unit 920 is configured to determine S business prediction results corresponding to S third sequences using a trained business prediction model; the S third sequences are obtained by respectively masking S groups of attribute fields from the original behavior sequence; a loss determination unit 930 is configured to determine S predicted losses corresponding to the S business prediction results based on the business tag; and an attribute determination unit 940 is configured to determine a key attribute field from the S groups of attribute fields based on the S predicted losses, for explaining the business prediction results of the original behavior sequence.

[0101] In one embodiment, any two groups of attribute fields in the S groups of attribute fields are mutually exclusive.

[0102] In one embodiment, the device 900 also includes a sequence generation unit 950, which is configured to replace the field values ​​of the attribute fields in the group corresponding to each behavior in the original behavior sequence with default values ​​for any group of attribute fields in the S groups of attribute fields, so as to obtain a corresponding third sequence.

[0103] In one embodiment, the business prediction model includes an attribute coding layer, a behavior representation layer, a sequence representation layer and an output layer; the result prediction unit 920 is specifically configured as follows: for each operation behavior in the T operation behaviors, using the attribute coding layer to process the field values ​​of the corresponding N attribute fields contained in the behavior to obtain the corresponding N attribute coding vectors; for any group of attribute fields in the S groups of attribute fields, setting the attribute coding vectors corresponding to the group of attribute fields in the N attribute coding vectors to zero; using the behavior representation layer, processing the N attribute coding vectors obtained after the zeroing process to obtain the behavior representation vector corresponding to the behavior; using the sequence representation layer, processing the T behavior representation vectors corresponding to the T operation behaviors to obtain the sequence representation vector of the third sequence corresponding to the group of attribute fields; using the output layer to process the sequence representation vector to obtain the business prediction result corresponding to the third sequence.

[0104] In one embodiment, the business prediction model includes an attribute coding layer, a behavior representation layer, a sequence representation layer and an output layer; the result prediction unit 920 is specifically configured as follows: for each operation behavior in the T operation behaviors, using the attribute coding layer to process the field values ​​of the corresponding N attribute fields contained in the behavior to obtain the corresponding N attribute coding vectors; for any group of attribute fields in the S groups of attribute fields, setting the model parameters in the behavior representation layer used to process the group of attribute fields to zero; using the behavior representation layer after the zeroing process, processing the N attribute coding vectors to obtain the behavior representation vector corresponding to the behavior; using the sequence representation layer, processing the T behavior representation vectors corresponding to the T operation behaviors to obtain the sequence representation vector of the third sequence corresponding to the group of attribute fields; using the output layer to process the sequence representation vector to obtain the business prediction result corresponding to the third sequence.

[0105] In one embodiment, the attribute determination unit 940 is specifically configured to: determine several predicted losses among the S predicted losses that are greater than a loss threshold or ranked within a preset front range; and determine several groups of attribute fields corresponding to the several predicted losses as the key attribute fields.

[0106] According to another embodiment, there is also provided a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute a combination of Figure 2 or Figure 5 The method described.

[0107] According to another embodiment of the present invention, a computing device is provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the Figure 2 or Figure 5 The method described.

[0108] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the present invention can be implemented by hardware, software, firmware or any combination thereof. When implemented by software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium.

[0109] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for enhancing the explainability of business results, comprising: Acquire an original behavior sequence and a corresponding service tag, wherein the original behavior sequence includes T operation behaviors performed by the target user and arranged in chronological order; Using the trained business prediction model, determining M business prediction results corresponding to M first sequences; the M first sequences are obtained by respectively shielding M groups of behaviors from the original behavior sequence; Based on the business tags, determining M prediction losses corresponding to the M business prediction results; Based on the M prediction losses, key behaviors are determined from the M groups of behaviors for explaining the business prediction results of the original behavior sequence.

2. The method according to claim 1, wherein: The business label is obtained through manual labeling.

3. The method according to claim 1, wherein: Get the original behavior sequence and the corresponding business tags, including: Obtaining the original behavior sequence; The business prediction model is used to determine the business prediction result of the original behavior sequence as the business label.

4. The method according to claim 1, wherein: The business tag includes multiple sub-tags corresponding to multiple prediction tasks, and the business prediction result includes multiple sub-results corresponding to the multiple prediction tasks.

5. The method according to claim 1, wherein: The business prediction model includes a behavior attention layer; wherein, before using the trained business prediction model to determine M business prediction results corresponding to the M first sequences, the method further includes: Obtaining T attention weights corresponding to the T operation behaviors determined by using the behavior attention layer; Determine some of the T attention weights that are greater than a weight threshold or are within a preset front range; The operation behaviors corresponding to the partial attention weights are grouped to obtain the M groups of behaviors.

6. The method according to claim 1 or 5, wherein: Any two groups of behaviors in the M groups of behaviors are mutually exclusive.

7. The method according to claim 1, wherein: Before using the trained business prediction model to determine the M business prediction results corresponding to the M first sequences, the method further includes: For any group of behaviors in the M groups of behaviors, the group of behaviors included in the original behavior sequence is replaced with default values ​​to obtain a corresponding first sequence.

8. The method according to claim 1, wherein: The business prediction model includes a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, M business prediction results corresponding to the M first sequences are determined, including: Determine T behavior representation vectors corresponding to the T operation behaviors by using the behavior representation layer; For any group of behaviors among the M groups of behaviors, setting the behavior representation vectors corresponding to the group of behaviors included in the T behavior representation vectors to zero; Using the sequence representation layer, the T behavior representation vectors obtained after the zeroing process are processed to obtain a sequence representation vector of the first sequence corresponding to the group of behaviors; The output layer is used to process the sequence representation vector to obtain a business prediction result corresponding to the first sequence.

9. The method according to claim 1, wherein: The business prediction model includes a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, M business prediction results corresponding to the M first sequences are determined, including: Determine T behavior representation vectors corresponding to the T operation behaviors by using the behavior representation layer; For any group of behaviors among the M groups of behaviors, setting the model parameters in the sequence representation layer for processing the group of behaviors to zero; Processing the T behavior representation vectors by using the sequence representation layer after the zeroing process to obtain a sequence representation vector of a first sequence corresponding to the group of behaviors; The output layer is used to process the sequence representation vector to obtain a business prediction result corresponding to the first sequence.

10. The method according to claim 1, wherein: Determining key behaviors from the M groups of behaviors based on the M predicted losses includes: Determine a number of predicted losses among the M predicted losses that are greater than a loss threshold or are ranked within a preset front range; A plurality of groups of behaviors corresponding to the plurality of predicted losses are determined as the key behaviors.

11. The method according to claim 10, wherein: Each operation behavior in the T operation behaviors corresponds to N attribute fields; wherein, after determining the key behavior from the M groups of behaviors based on the M predicted losses, the method further includes: Determine Q business prediction results corresponding to Q second sequences using the business prediction model; the Q second sequences correspond to sequences obtained by respectively masking Q groups of attribute fields in a behavior sequence formed by the key behaviors; Based on the business label, determining Q prediction losses corresponding to the Q business prediction results; Based on the Q prediction losses, a key attribute field is determined from the Q groups of attribute fields, for explaining the business prediction result of the original behavior sequence.

12. A method for enhancing the explainability of business results, comprising: Acquire an original behavior sequence and a corresponding service tag, wherein the original behavior sequence includes T operation behaviors performed by the target user and arranged in chronological order, wherein each operation behavior includes N attribute fields; Determine S business prediction results corresponding to S third sequences by using the trained business prediction model; the S third sequences are obtained by respectively masking S groups of attribute fields from the original behavior sequence; Determining, based on the service label, S prediction losses corresponding to the S service prediction results; Based on the S prediction losses, a key attribute field is determined from the S groups of attribute fields, for explaining the business prediction result of the original behavior sequence.

13. The method according to claim 12, wherein: Any two groups of attribute fields in the S groups of attribute fields are mutually exclusive.

14. The method according to claim 12, wherein: Before using the trained business prediction model to determine the S business prediction results corresponding to the S third sequences, the method further includes: For any group of attribute fields in the S groups of attribute fields, the field values ​​of the attribute fields in the group corresponding to each behavior in the original behavior sequence are replaced with default values ​​to obtain a corresponding third sequence.

15. The method according to claim 12, wherein: The business prediction model includes an attribute encoding layer, a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, S business prediction results corresponding to S third sequences are determined, including: For each operation behavior in the T operation behaviors, using the attribute coding layer to process field values ​​of corresponding N attribute fields contained in the behavior to obtain corresponding N attribute coding vectors; For any group of attribute fields in the S groups of attribute fields, setting the attribute coding vectors corresponding to the group of attribute fields in the N attribute coding vectors to zero; Using the behavior representation layer, the N attribute encoding vectors obtained after the zeroing process are processed to obtain the behavior representation vector corresponding to the behavior; Using the sequence representation layer, processing the T behavior representation vectors corresponding to the T operation behaviors to obtain a sequence representation vector of a third sequence corresponding to the group of attribute fields; The output layer is used to process the sequence representation vector to obtain a business prediction result corresponding to the third sequence.

16. The method according to claim 12, wherein: The business prediction model includes an attribute encoding layer, a behavior representation layer, a sequence representation layer and an output layer; wherein, using the trained business prediction model, S business prediction results corresponding to S third sequences are determined, including: For each operation behavior in the T operation behaviors, using the attribute coding layer to process field values ​​of corresponding N attribute fields contained in the behavior to obtain corresponding N attribute coding vectors; For any group of attribute fields in the S groups of attribute fields, setting the model parameters used for processing the group of attribute fields in the behavior representation layer to zero; Using the behavior representation layer after the zeroing process, the N attribute encoding vectors are processed to obtain a behavior representation vector corresponding to the behavior; Using the sequence representation layer, processing the T behavior representation vectors corresponding to the T operation behaviors to obtain a sequence representation vector of a third sequence corresponding to the group of attribute fields; The output layer is used to process the sequence representation vector to obtain a business prediction result corresponding to the third sequence.

17. The method according to claim 12, wherein: Determining a key attribute field from the S groups of attribute fields based on the S prediction losses includes: Determine a number of predicted losses among the S predicted losses that are greater than a loss threshold or are ranked within a preset front range; Several groups of attribute fields corresponding to the several predicted losses are determined as the key attribute fields.

18. A device for enhancing the interpretability of business results, comprising: A sequence acquisition unit, configured to acquire an original behavior sequence and a corresponding service tag, wherein the original behavior sequence includes T operation behaviors performed by a target user and arranged in chronological order; A result prediction unit is configured to determine M business prediction results corresponding to M first sequences using a trained business prediction model; the M first sequences are obtained by respectively shielding M groups of behaviors from the original behavior sequence; a loss determination unit configured to determine, based on the service tag, M predicted losses corresponding to the M service prediction results; A behavior determination unit is configured to determine key behaviors from the M groups of behaviors based on the M predicted losses, so as to explain the business prediction results of the original behavior sequence.

19. A device for enhancing the interpretability of business results, comprising: A sequence acquisition unit is configured to acquire an original behavior sequence and a corresponding service tag, wherein the original behavior sequence includes T operation behaviors made by the target user arranged in chronological order, wherein each operation behavior includes N attribute fields; A result prediction unit is configured to use the trained business prediction model to determine S business prediction results corresponding to S third sequences; the S third sequences are obtained by respectively masking S groups of attribute fields from the original behavior sequence; a loss determination unit, configured to determine S predicted losses corresponding to the S business prediction results based on the business tag; The attribute determination unit is configured to determine a key attribute field from the S groups of attribute fields based on the S prediction losses, so as to explain the business prediction result of the original behavior sequence.

20. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed in a computer, the computer is caused to execute the method according to any one of claims 1 to 17.

21. A computing device comprising a memory and a processor, wherein: The memory stores executable codes, and when the processor executes the executable codes, the method according to any one of claims 1 to 17 is implemented.

Images