WO2026001385A1 - Information processing methods, store diagnosis method, intelligent assistant working method, system and device - Google Patents

Abstract

Provided in the embodiments of the present disclosure are information processing methods, a store diagnosis method, an intelligent assistant working method, a system and a device. A solution provided in the present disclosure uses a causal inference algorithm to build a text repository (for example, at least one factor of a preset question stored in the text repository, which is obtained by performing attribution analysis on the preset question by using the causal inference algorithm), and empowers a merchant-side management tool with a first preset model which is fine-tuned by using a training sample set related to a store operation scenario, so as to upgrade the level of intelligence of the management tool. The text repository provides data support to retrieve a store-diagnosis result and/or at least one factor of a user's query question, as well as suggestion content linked to said factor; the first preset model provides intelligent response support; and, on the basis of the store diagnosis result and/or the user's query question and the suggestion content linked to the at least one factor, the first preset model can output a text that better conforms to natural language logic and is suitable for the store operation scenario.

Description

Information processing, store diagnostics, and intelligent assistant working methods, systems, and equipment

This disclosure claims priority to Chinese Patent Application No. 202410867589.9, filed with the China Patent Office on June 28, 2024, entitled “Information Processing, Store Diagnosis and Intelligent Assistant Working Method, System and Device”, the entire contents of which are incorporated herein by reference.

Technical Field

This disclosure relates to the field of computer technology, and in particular to a method, system and device for information processing, store diagnosis and intelligent assistant operation.

Background Technology

There are many merchants on e-commerce platforms. Merchants can manage their stores on the e-commerce platform through the management tools provided by the platform. For example, they can use the management tools to obtain recent traffic data of the store (such as the number of visits, the number of orders, the number of paid orders, etc.) and set keywords for the store's products to improve the chances of product search.

However, existing management tools are not intelligent enough; merchants can only access some statistical data about their stores through these tools. Experienced merchants can use this data to pinpoint store problems and adjust their business strategies based on their extensive experience. But for less experienced merchants, these statistics are insufficient to accurately identify store issues, and they don't know how to adjust their business strategies.

Summary of the Invention

The embodiments of this disclosure provide an information processing, store diagnosis, and intelligent assistant working method, system, and device to help merchants operate their stores.

In a first embodiment of this disclosure, an information processing method is provided. The method includes:

For the target store, identify the target problem;

Retrieve at least one factor of the target question from the document database;

Based on the target question and the suggested content associated with at least one factor, a prompt word is generated.

The first preset model is used to output the text corresponding to the Prompt, so as to help users solve or improve the target problem;

The document database stores at least one factor of the preset question obtained by attribution analysis of the preset question using a causal inference algorithm; the first preset model is obtained by adjusting and training the pre-trained model based on the pre-trained model for intelligent response tasks in the store operation scenario.

In a second embodiment of this disclosure, an information processing method is provided. The method includes:

Construct a query template library, which includes multiple query templates, and each query template is associated with a preset question related to store operations;

Based on the Query template library, a copywriting database is constructed; wherein, the copywriting database stores at least one factor of the Query template obtained by attribution analysis of preset questions associated with the Query template using a causal inference algorithm;

A training sample set is generated based on the Query template library and the copywriting database;

Using the training sample set, the pre-trained model is adjusted and trained to obtain a first preset model that can handle intelligent response tasks in store operation scenarios.

In a third embodiment of this disclosure, a store diagnostic method is provided. This method is applicable to the client side, and specifically, the method includes:

Respond to diagnostic trigger events for the target store, trigger store diagnostics to present and/or play the diagnosed target issues;

Display and/or play text, which includes suggestions to help users solve or improve the target problem;

The copy is obtained using a first preset model, which is obtained by adjusting and training the pre-trained model based on a pre-trained model for intelligent response tasks in a store operation scenario; the suggested content is associated with at least one factor of the target problem, and the at least one factor of the problem is obtained by attribution analysis of the target problem using a causal inference algorithm.

In the fourth embodiment of this disclosure, a method for operating a smart store assistant is provided. This method is applicable to the server side, and specifically, the method includes:

Display the target store's management page;

Monitor user actions on the management page to identify user-defined issues related to store operations;

In response to a user's confirmation command triggered for the target problem, display and/or play text, which includes suggestions to help the user solve or improve the target problem;

The copy is obtained using a first preset model, which is obtained by adjusting and training the pre-trained model based on a pre-trained model for intelligent response tasks in a store operation scenario; the suggested content is associated with at least one factor of the target problem, and the at least one factor of the problem is obtained by attribution analysis of the target problem using a causal inference algorithm.

In a fifth embodiment of this disclosure, a service system is provided. This service system includes a client and a server. The client is used to implement the steps in the above-described store diagnosis method embodiment or the steps in the above-described store intelligent assistant working method embodiment. The server is used to implement the steps in the above-described information processing method embodiments.

In a sixth embodiment of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor. The memory stores executable instructions; the processor executes the executable instructions to perform the steps as described in the above method embodiments.

In a sixth embodiment of this disclosure, a computer-readable storage medium is provided. This storage medium stores computer instructions that, when executed by a processor, can perform the steps described in the above method embodiments.

A seventh embodiment of this disclosure also provides a computer program product comprising a computer program or instructions that, when executed by a processor, cause the processor to perform the steps in the above method embodiments.

The pre-trained models mentioned in the embodiments of this disclosure refer to machine learning models with a large number of parameters generated by training with massive amounts of data using self-supervised methods. They provide excellent distributed feature representation capabilities and model generalization capabilities for downstream tasks (such as the intelligent response task in the store operation scenario mentioned in the embodiments of this disclosure). Pre-trained models can also be called generalized large models.

The technical solution provided in this disclosure combines a large model with causal inference technology. Specifically, a causal inference algorithm is used when constructing the copywriting database to perform attribution analysis on the preset questions associated with the query template to obtain at least one factor. The training samples used to adjust and train the pre-trained model are generated based on the query template library and the copywriting database. Using a training sample set composed of multiple such training samples, the pre-trained model is adjusted and trained to adapt it into a first preset model suitable for store operation scenarios.

The solution provided in this disclosure utilizes a causal inference algorithm to construct a document database (such as at least one factor of a preset question obtained by attribution analysis of a preset question using the causal inference algorithm stored in the document database). A first preset model, adjusted using a training sample set related to store operation scenarios, empowers the merchant-side management tool, thereby enhancing the intelligence of the management tool. The document database and the first preset model can be deployed on the server side, and the corresponding product form on the client side can be a smart assistant. Users can use this smart assistant to diagnose target stores or ask questions related to store operations. The document database provides data support for the smart assistant, retrieving at least one factor of the store diagnosis results and/or the user's question, as well as suggested content associated with that factor. The adjusted and trained first preset model provides intelligent response support for the smart assistant. Based on the store diagnosis results and/or the user's question, and suggested content associated with at least one factor, the first preset model can output document that is more consistent with natural language logic and suitable for store operation scenarios, thus assisting users in operating their stores.

As can be seen, the technical solutions provided in the various embodiments of this disclosure utilize large models and causal inference techniques to enhance the platform's intelligent service capabilities and assist merchants in operating their stores.

Attached Figure Description

To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

Figure 1 is a schematic diagram of the technical architecture corresponding to the technical solution provided in the embodiments of this disclosure;

Figure 2 is a schematic diagram of a service system provided in an embodiment of this disclosure;

Figure 3 is a flowchart illustrating an information processing method provided in an embodiment of this disclosure;

Figure 4 is a flowchart illustrating an information processing method provided in another embodiment of this disclosure;

Figure 5 is a schematic flowchart of a store diagnosis method provided in an embodiment of this disclosure;

Figure 6a is a schematic diagram of the first interface corresponding to the technical solution provided in an embodiment of this disclosure;

Figure 6b is a schematic diagram of the second interface corresponding to the technical solution provided in an embodiment of this disclosure;

Figure 6c is a schematic diagram of a third interface corresponding to a technical solution provided in an embodiment of this disclosure;

Figure 7 is a flowchart illustrating the working method of a smart store assistant provided in an embodiment of this disclosure;

Figure 8 is a structural block diagram of an electronic device provided in an embodiment of this disclosure.

Detailed Implementation

To enable those skilled in the art to better understand the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

In some processes described in the specification, claims, and accompanying drawings of this disclosure, multiple operations appearing in a specific order are included. These operations may be executed out of order or in parallel. Operation numbers such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the terms "first," "second," etc., used herein are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types. The embodiments described below are merely some embodiments of this disclosure, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

Before introducing the various embodiments, a brief introduction to some technical terms used in this document will be given.

Query: A query is a message sent by a search engine or database in order to find a file, website, record, or series of records in a database.

Pre-trained models refer to machine learning models with a large number of parameters generated by training on massive amounts of data using self-supervised methods. They provide excellent distributed feature representation capabilities and model generalization capabilities for downstream tasks. Pre-trained models can be called generalized large models.

SFT: Supervised Fine-tuning. This involves fine-tuning a pre-trained model based on a large pre-trained model for a specific task in a particular scenario. During fine-tuning, the pre-trained model's parameters and structure are adjusted according to the characteristics of the task to improve its performance on that task.

LoRA: A model fine-tuning method that freezes the model parameters of a pre-trained model and injects trainable low-rank matrices into the attention layer of a Transformer architecture (a model that uses attention mechanisms to improve model training speed), which greatly reduces the number of trainable parameters for downstream tasks and improves training speed.

SHAP: Shapley Value, a method for calculating the causality between features and labels. It calculates the marginal benefit of adding a feature to a given feature combination relative to not adding that feature by enumerating different feature combinations.

Tars: ICUB large-scale language model, a deep learning model trained on massive amounts of text data.

ab: Inquiry, which is a question made by a buyer or seller to the other party regarding the terms of a transaction in order to buy or sell a certain product.

rfq: Request for quote.

To facilitate understanding of the technical solutions, the embodiments of this disclosure are described from the perspective of the underlying technical architecture. The technical solutions provided in the following embodiments are all implemented based on this technical architecture. Refer to Figure 1, which shows the technical architecture diagram of the technical solutions provided in this disclosure. As shown in Figure 1, the technical architecture of the technical solutions provided in this disclosure includes:

I. Building a Query Template Library (Figure 1, number 101)

In one feasible technical solution, building a query template library may include:

S11. Determine multiple basic query templates;

S12. Using the second preset model, the multiple basic query templates are extended to obtain multiple extended query templates;

S13. Evaluate the multiple extended query templates and retain the extended query templates that pass the evaluation;

The Query template library includes multiple basic Query templates and multiple extended Query templates that have passed evaluation.

For example, the multiple basic query templates in S11 mentioned above may include, but are not limited to: multiple basic query templates covering all traffic nodes, multiple basic query templates related to store optimization items, multiple basic query templates related to store operation issues, and so on. In e-commerce scenarios, traffic nodes may include, but are not limited to: keyword search, user visits, adding to cart, inquiries, order placement, payment, etc. Store optimization items may include, but are not limited to: in-store product keywords, in-store product images, store promotional products, target audience for products, etc. Store operation issues may include, but are not limited to: how to increase search visibility, how to increase payment volume, how to increase store traffic, etc. The stores mentioned in this article can be online stores on e-commerce platforms or online stores on social media platforms, etc. Online stores can also be called "online stores" or "e-commerce shops." E-commerce platforms or social media platforms provide merchants with technical support for operating stores on the internet, displaying and selling goods or services in online stores. E-commerce platforms or social media platforms utilize internet technology to allow consumers to browse product information, compare, place orders, pay for goods, and wait for delivery to a designated address.

Among them, the multiple basic query templates covering all traffic nodes can be divided into two main categories. The first category consists of multiple basic query templates corresponding to differences between the store's own traffic data and the traffic data of reference stores (such as peer stores and/or stores of the same level). The second category consists of multiple basic query templates corresponding to differences between the store's traffic data in the first time period and the historical traffic data in the second time period.

After establishing multiple basic query templates, a second preset model can be used to offline expand multiple extended query templates. The second preset model can be a Large Language Model (LLM), a deep learning model trained on massive amounts of text data. It can not only generate natural language text but also deeply understand the meaning of text and handle various natural language tasks.

Multiple extended queries derived from the second preset model may not all meet the requirements. Therefore, the multiple extended queries output by the second preset model can be evaluated. The evaluation process may include, but is not limited to: obtaining evaluation rules; evaluating each extended query separately using the evaluation rules, and retaining the extended query templates that pass the evaluation. Furthermore, a human intervention interface can be added to the evaluation process, such as displaying the extended query templates that pass the evaluation on the client interface, and allowing users to edit, delete, etc., the extended query templates that pass the evaluation.

This disclosure does not specify the evaluation rules, and may include multiple rule items, each of which can be set manually.

Finally, a query template library is obtained, containing multiple basic query templates and multiple extended query templates that have passed evaluation. During store diagnostics, the query templates in the library can be sorted based on the store's diagnostic results (such as low store exposure, low click-through rate, low inquiry volume, etc.). For example, query templates closely related to the store's diagnostic results can be ranked higher. This allows the top-ranked templates (e.g., 1, 2, 3, or 4) to be used as the corresponding query templates for the store's diagnostic results. Alternatively, through big data statistics, common problems found in stores of the same industry or level can be identified, and the query templates in the library can be sorted based on these statistical results. This allows the top-ranked templates (e.g., 1, 2, 3, or 4) to be used as recommended query templates.

In summary, the Query template library may include at least one of the following:

The query template for the reason why the data of any store on the platform differs from that of a reference store;

Query templates for explaining why data discrepancies exist for any store on the platform at different times;

Query templates for store management issues;

Query templates corresponding to store optimization items;

The query template is extended using the second preset model.

II. Building a document database (Figure 1, number 102)

2.1 Collect basic data, such as store attribute information, store traffic data, and store operational data on the platform. Store attribute information may include, but is not limited to: the store's industry, store rating, etc. Store traffic data may include, but is not limited to: historical exposure rate, click-through rate, inquiry volume, and average daily order volume for the past 30 days, 60 days, etc. Store operational data may include, but is not limited to: number of store showcases, number of featured products, number of self-marketing interactions, RFD amount, advertising expenditure, and store traffic generation (including the number of visitors entering the store through different channels such as search, events, paid advertising, and live streaming), etc.

2.2 Attribution of Store Optimization Items

Causal inference is a statistical and reasoning method aimed at understanding and inferring causal relationships between events, variables, or behaviors. This disclosed solution utilizes causal inference techniques to analyze abnormal traffic nodes that cause store problems; these traffic nodes represent potential store optimization opportunities. Funnel analysis is employed in the attribution analysis. Specifically, funnel analysis is used to analyze the store's traffic and operational data on the platform to identify store optimization options for resolving or improving store issues.

Data funnel nodes are a key element in funnel analysis, representing the various traffic points encountered from the start to the end of an event. These nodes include the starting point, intermediate process nodes, and the ending point. Specifically, the data funnel node settings will vary depending on the application scenario. In e-commerce, data funnel nodes may include: customer visits, adding to cart, placing orders, payment, etc. By analyzing the data from these nodes, we can understand the customer's conversion rate at each stage, thereby identifying potential problems and optimization areas. For example, finding a low conversion rate at a certain node may indicate a problem at that node, requiring further adjustments and optimizations. This node represents the store's optimization needs.

Therefore, when performing funnel analysis, it is necessary to carefully identify and define the nodes of the data funnel to ensure the accuracy and effectiveness of the analysis. This disclosure does not specifically limit the identification and definition of the nodes of the data funnel; each scenario requires specific analysis. In practice, they can be determined and defined based on experience.

When performing funnel analysis, the monotonic stack method can be used to perform attribution analysis of node optimization points in the flow funnel. The monotonic stack is often used to find the position of the first element to the left or right of a given element in an array that is larger or smaller than it, thereby identifying local maxima and minima in the sequence.

For example, in practical implementation, two monotonically increasing stacks can be maintained to find the node where the metrics (visits, inquiries, etc.) related to the questions asked by merchants or the problems diagnosed in the merchant's store differ the most from those of a reference store (another store in the same industry or at the same level as the merchant's store). This is essentially the position of the minimum value at each position in an array. By using two monotonically increasing stacks, the location of the abnormal node can be effectively found; this position is the key point for optimization, i.e., the store optimization item.

2.3 Factor Ranking of Store Optimization Items

Modeling is performed with store optimization as the target, constructing a causal tree model based on multiple factors. This causal tree model can be a LightGBM model (LGB model for short). In practice, the multiple factors can be selected from multiple search ranking features in e-commerce search scenarios. The causal tree model can be trained based on a pre-set training set, which can be constructed based on store data from the platform. Factors that cannot be changed, such as store type and statistical features, will be removed during training to avoid interfering with the splitting nodes of the causal tree model and affecting the feature importance effect.

A dataset is constructed based on store data (such as store traffic data and store operational data) from stores on the platform. This dataset differs from the training set mentioned above; it can be referred to as the prediction set. After the causal tree model is trained on the training set, the dataset is used to calculate the influence of each of the multiple factors on the prediction results of the causal tree model, thus obtaining the influence degree of each of the multiple factors.

Specifically, the calculated influence can be expressed as the SHAP value (Shapley value). SHAP (Shapley Additive exPlanations) is a "model interpretation" package that can interpret the output of any machine learning model. All features are considered "contributors." For each predicted sample, the model produces a predicted value, and the SHAP value is the numerical value assigned to each feature in that sample. In this disclosure, the SHAP value represents the influence or contribution of each factor to the predicted result of the causal tree model output. The overall idea for calculating the SHAP value is to enumerate different factor combinations and calculate the impact of adding a factor relative to not adding that factor on the predicted result under a certain factor combination. This impact is the marginal benefit of that factor for that factor combination, as shown in the following formula:

Where V(i) represents the Shapley value of factor i among multiple factors; N represents the set of multiple factors; |N| represents the number of multiple factors; S is a subset of N; |S| is the number of subsets of N; v(S) represents the value of multiple factors in set S.

In the calculation of the Shapley value, all possible combinations of factors S are traversed, and the additional contribution of factor i after being added to combination S is calculated, i.e., v(S∪{i})-v(S). Then, a weighted average of the additional contributions of all combinations is calculated, where the weights are determined according to the size of the combination.

The above formula means that the Shapley value equals the average contribution of a factor across all possible combinations of cooperation. It considers the marginal contribution of each factor and determines the final allocation through a weighted average. The Shapley value of a factor is defined as its contribution to the prediction results of the causal tree model. When there are many factors, approximate values for the Shapley value can be calculated using methods such as factor sampling or the assumption of factor independence.

Because the preceding modeling focused on store optimization items, a causal tree model was constructed based on multiple factors; then, the SHAP value of each factor was calculated using the aforementioned algorithm. Therefore, after obtaining the SHAP values of multiple factors, they can be sorted according to their SHAP values; thus, the factor ranking of the store optimization items can be obtained.

There are many aspects to store optimization, such as whether specific channels are used to improve click-through rate, and whether keywords are accurate. These are all store optimization items. Using the techniques described above, the factors for each store optimization item can be ranked.

In fact, stores belonging to different industries, at different levels (as determined by the platform's rating rules), and at different stages (such as the growth stage or the stage of preparing for level upgrades) will have different operational approaches. In other words, the factor ranking for the same optimization item will differ depending on the industry, level, or stage of the store. For example, stores on the platform can be grouped based on their primary or secondary industry, level, whether they advertise, whether they are in the growth stage, or whether they are in the stage of preparing for level upgrades.

Primary and secondary industries are two levels of industry classification, forming a classification system that ranges from broad to specific. Primary industries are the most basic level of industry classification and can include agriculture, forestry, animal husbandry, fishery, industry, construction, etc. Secondary industries are further subdivided from these primary industries into broader industry categories.

Therefore, this disclosure proposes the concept of grouping stores on the platform, and calculating the factor ranking of corresponding store optimization items for stores in different groups. That is, in the above technical solution, the dataset is constructed based on store data belonging to the same group (such as store traffic data and store operational data). Then, using the dataset constructed from store data within the same group, the SHAP values of each factor are calculated, and the factor ranking obtained based on the SHAP values is the factor ranking of the store optimization items for that group. Simply put, using the above technical solution, for a unified store optimization item, the factor ranking of the store optimization item corresponding to different groups will be calculated. For ease of understanding, this can be represented as shown in Table 1 below:

2.4 Answer Database

Each factor is associated with suggested content, which can be a task that the store can perform (such as changing keywords, running ads, or using auxiliary channels). Furthermore, different copy templates can be built for different query templates. For example, for service, advertising, or transaction tasks, the copy template can specify that in addition to suggested content, it should also provide a comparison of data between the target store and the reference store, and so on. For product or sub-account tasks, the copy template can specify that in addition to suggested content, it should also provide a list of optimization targets (target products or sub-accounts, etc.), and so on.

In summary, the copywriting database can store: attribute information of stores on the platform, traffic and operational data of stores on the platform, store optimization items associated with query templates, factor sorting of store optimization items, suggested content associated with factors, and a corpus knowledge base of copywriting template information corresponding to different query templates.

III. Model Deployment (Figure 1, No. 103)

3.1 Generating training samples

Based on the text database obtained in Part (II) and the query template database obtained in Part (I), a third preset model is invoked to generate multiple training samples. The third preset model can be a large language model (LLM), etc., but this embodiment does not specify a particular model. For example, the third preset model can be invoked to generate 4000 or more (e.g., 5000, 6000, etc.) valid data offline as training samples. Some training samples contain sensitive data (such as merchant privacy information), which needs to be anonymized, for example, by using a mask. Simultaneously, the prompt words can be adjusted to ensure that the text output by the first preset model is in a specified format, such as JSON (JavaScript Object Notation, a lightweight data exchange format), facilitating subsequent integration between groups and jumps.

For example, the Prompt explicitly states that the model is expected to return the text in JSON format. Here's a simple JSON example: An example JSON output might look like this: {"Question":"What is the target question?","Answer":"This is the corresponding text."} Please provide your answer in this format.

3.2 Model Adjustment and Training

Using the training samples generated in section 3.1 above, a fine-tuning method (such as LoRA) can be used to train a pre-trained model to enhance the logic and accuracy of the pre-trained model's output in store operation scenarios. The advantage of LoRA is that it only trains a small number of parameters of the pre-trained model, and improves the model's prediction accuracy by adding a low-rank matrix to the attention layer of the Transformer.

The pre-trained model can be a large language model (LLM). Pre-trained models learn rich feature representations and pattern recognition capabilities. Through fine-tuning, these learned feature representations can be used to quickly and effectively adapt to specific tasks (such as tasks in a store operation scenario). Fine-tuning is faster than training a model from scratch, thus accelerating the training process.

IV. Online Link Setup (104 in Figure 1)

In practical implementation, online deployment can be completed using Rlab (a high-level programming language for engineering application computing). Of course, other tools can also be used for online deployment; this disclosure does not specifically limit this, but only exemplifies the use of Rlab. The online infrastructure also requires support from, but is not limited to, EAS, Igraph, pre-trained model development tools (Tars Studio), PAI, and Itag tools. EAS (Elastic Algorithm Service) is an online model service provided by the machine learning platform product for online inference scenarios to achieve one-stop model development and deployment applications. It supports deploying model services in public or dedicated resource groups, enabling real-time response to model loading and data requests based on heterogeneous hardware (such as CPUs or GPUs). Igraph is an online graph storage and query system that provides storage, query, update, and computation services for large-scale graph data. The pre-trained model development tools (Tars Studio) provide visual development tools. PAI is a machine learning platform designed to enable developers to use artificial intelligence (AI) technologies more efficiently, simply, and in a standardized manner. Itag provides data annotation services to help users complete data annotation tasks quickly and efficiently.

Online connections may include, but are not limited to:

S21. Determine the target query.

S22, Rewrite the target query.

The rewriting of the target query may include, but is not limited to:

Query unification: Rewrite the target query through simple string manipulation, such as removing spaces, case conversion, and handling special characters;

Query correction: Proactively corrects target queries that contain spelling errors;

Query expansion: Expand the target query based on its semantics or intent to find a set of queries that are consistent with the target query and retrieve them together with the target query;

etc.

S23. Obtain data based on Igraph.

The data obtained may include: data retrieved from a copywriting database. This data may include, but is not limited to: target store data (such as store traffic and operational data), reference store data (such as reference store traffic and operational data), and copywriting information. Copywriting information may include, but is not limited to: copywriting templates, query templates matching the target query after the above S22 rewriting process, store optimization items associated with the matched query, at least one factor of the store optimization item, suggested content associated with the factor, etc.

S24. Preprocess the acquired data.

This disclosure does not specifically limit the term "preprocessing," which can include, but is not limited to, operations such as data cleaning, standardization, and format conversion. For example, data cleaning can be used to filter useful data from the acquired data. For instance, unstructured data (such as text) may require structuring processes such as word segmentation and stemming. Based on the input requirements of the pre-trained model, the data needs to be converted into an appropriate format, such as JSON or text format.

S25. Prompt prompt concatenation

The matched query template can be concatenated with the preprocessed data described above into the Prompt. The output (or response) quality of the first preset model (such as a finely tuned large language model) depends on the prompt word (i.e., the Prompt). Simply put, the Prompt is the question the user wants to ask, which is sent as input to the first preset model. The first preset model attempts to understand this input and then outputs an appropriate answer or response. Concatenating the Prompt here makes the first preset model output more accurate and useful copy.

The Prompt is a crucial component, guiding the model's behavior. Adjusting the Prompt can improve the quality of the model's output. In practice, designers can manually design Prompt splicing rules to ensure that the splicing process adheres to these preset rules during Prompt splicing.

V. Front-end Function Deployment

With the technical architecture support of the above parts one, two, three and four, the corresponding functional interfaces can be deployed on the front-end application (APP) or browser webpage, so that users can call the corresponding services in the back-end (such as the server with the above technical architecture) through the functional interfaces, such as store diagnosis service, intelligent question answering service, etc.

The functional interface can take the form of functional controls on the app page or browser webpage, allowing users to invoke corresponding backend services by clicking on these controls. Alternatively, the app page or browser webpage can display a specific icon, such as a smart assistant icon. This icon can be any pattern or animation, without limitation. Users can activate the smart assistant function via voice or gesture, and the smart assistant icon will change its display style to indicate successful activation. Users can then interact with the smart assistant to invoke corresponding backend services. In short, the product form presented on the client side of the above technical architecture can be the smart assistant icon.

The store diagnostic service may include: diagnosing the store and, when problems are identified, providing optimization suggestions strongly related to those problems. Furthermore, the service may also offer action recommendations for improving these optimizations.

Intelligent question-answering services may include: responding to user-inputted behavioral suggestion questions, and outputting the corresponding action strategies and goals for those behavioral suggestion questions, etc.

In summary, the technical solution provided in this disclosure has at least the following technical effects:

1. The technical solution provided in this disclosure can rewrite and polish the output copy by adjusting and training the pre-trained model to suit downstream tasks (such as intelligent response tasks in store operation scenarios), thereby achieving content personalization.

2. Starting with multiple factors, establish causal tree models for different store optimization items, thereby determining the factor ranking of different store optimization items. Factor associations include suggested content, such as suggested tasks that merchants can perform.

3. Because there are differences between different industries and stores of different levels, the factor ranking for the same optimization item will differ for stores in different industries (e.g., how to increase traffic). For example, two stores in the machinery and equipment industry and two stores in the clothing industry will have different factor rankings when addressing the issue of increasing traffic. Therefore, this disclosed technical solution groups the stores on the platform; when determining the factor ranking for different store optimization items, the factor ranking for optimization items under different groups is determined separately.

The technical solution disclosed herein combines a pre-trained model with causal inference technology to achieve practical application in e-commerce scenarios. Specifically, this solution constructs a query template library suitable for store operation scenarios; it also constructs a copywriting resource library suitable for store operation scenarios based on the query template library. Furthermore, when constructing the copywriting resource library, causal inference technology is used to analyze attribution and identify at least one factor for each store operation-related issue, where the factor is associated with suggested content. In this way, a training sample set can be generated based on the query template library and the copywriting resource library, and this training sample set can be used to adjust and train the pre-trained model to suit store operation scenarios, enhancing the logic and accuracy of the output copywriting in store operation scenarios.

The technical solutions corresponding to the technical architecture mentioned above (as shown in parts one, two, three, and four of Figure 1) can be deployed on the server side. This server side can be a server, service cluster, virtual server, or cloud, etc., and this embodiment does not specifically limit this. The server side provides corresponding functional services to the client side, such as store diagnostics, intelligent Q&A services, etc. Users can access the management page provided by the platform through a browser or client application (APP) on their client devices. Client devices can be, but are not limited to: smartphones, smart wearable devices, tablets, laptops, desktop computers, etc. As shown in Figure 2, one embodiment of this disclosure provides a service system, which includes a server side and a client side.

Server 11 is used to construct a query template library, which includes multiple query templates, each associated with a preset question related to store operations. Based on the query template library, a copywriting database is constructed. The copywriting database stores at least one factor of the query template obtained by attribution analysis of the preset question associated with the query template using a causal inference algorithm. A training sample set is generated based on the query template library and the copywriting database. The pre-trained model is adjusted and trained using the training sample set to obtain a first preset model capable of handling intelligent response tasks in store operation scenarios.

Client 12 is used to deploy the corresponding interface so that users can call the server through the interface to retrieve target copywriting materials in the copywriting material database and output the copywriting corresponding to the copywriting materials using the first preset model.

Furthermore, the service system, including server 11 and client 12, is also used to implement the following functions. Specifically:

Client 12 is also configured to respond to diagnostic trigger events for a target store, trigger store diagnostics to present and/or play the diagnosed target problem, and send information to the server regarding the target problem; or, display the management page of the target store; listen to the user's operations on the management page to capture the user's target problems in store operations; respond to the user's confirmation instruction triggered for the target problem, and send information regarding the target problem to the server.

Server 11 is also used to identify target problems for target stores; retrieve at least one factor of the target problem from the copywriting database; generate a prompt word based on the target problem and the suggested content associated with the at least one factor; and output the copywriting corresponding to the prompt using a first preset model to help users solve or improve the target problem.

Client 12 is also used to receive the text fed back by the server, and to display and/or play the text.

It's important to note that the "store diagnosis" described above can be performed by client 12. After triggering the diagnosis, client 12 can obtain store data from the reference store and/or historical store data of the target store from server 11, etc., and then call its local diagnostic program to perform a store diagnosis on the target store. Alternatively, the "store diagnosis" can be performed by server 11. After client 12 triggers the diagnosis, it sends a target store diagnosis request to server 11 so that server 11 can obtain the target store's current store data, historical store data, reference store data, etc., and perform a diagnosis based on the obtained data to identify the target problem. Server 11 then returns the identified target store to client 12.

As can be seen, the service system provided in this embodiment utilizes the powerful computing power of the server 11 to build a query template library and use causal inference technology to find at least one factor that causes store problems; it also uses a large model (i.e., the first preset model) to realize the logic and accuracy of the output copy in the store operation scenario, so as to help users (or store merchants) grow.

It is important to note here that the above description uses a store operation scenario as an example to introduce the technical solution disclosed herein. In fact, the technical architecture provided in this disclosure can also be applied to other specific scenarios, such as enterprise management scenarios, project management scenarios, and so on.

The technical solutions provided in this disclosure will be described below by way of method embodiments.

Figure 3 shows a flowchart illustrating an information processing method provided in an embodiment of this disclosure. The executing entity of the method provided in this embodiment can be a server in the aforementioned service system. As shown in Figure 3, the method may include:

301. Identify the target problem for the target store.

302. Retrieve at least one factor of the target question from the document database.

303. Based on the target question and the suggested content associated with the at least one factor, generate the prompt word Prompt.

304. Utilize the first preset model to output the text corresponding to the Prompt, in order to help users solve or improve the target problem.

The document database stores at least one factor of the preset question obtained by attribution analysis of the preset question using a causal inference algorithm; the first preset model is obtained by adjusting and training the pre-trained model based on the pre-trained model for intelligent response tasks in the store operation scenario.

The technical architecture described above provides an example of building a query template library. In fact, a query template corresponds to a question. That is, a query template can be directly understood as a corresponding question template. Alternatively, in practical implementation, building a query template library can also be interpreted as "building a question library."

When determining the target problem of a target store, the aforementioned 301 can do so by diagnosing the target store or by determining the target problem based on user actions. Determining the target problem based on user input may include, but is not limited to: determining the target problem based on user input information, determining the target problem based on a user request triggered by a setting item, or having the network side recommend at least one recommended problem based on big data analysis, from which the user selects the recommended problem as the target problem.

That is, the aforementioned 301 "Identify the target issue for the target store" may include:

3011. Responding to user input, determining a target question for the target store based on the user input information; and/or

3012. Responding to a user's request triggered on the management page of a target store for a setting item, and for said setting item, determining the target problem; and/or

3013. Diagnose the target store and determine the target problem based on the diagnosis results; and/or

3014. Obtain at least one recommendation question recommended by the network side based on big data analysis, and respond to the selection event by taking the selected recommendation question as the target question.

In one specific implementation, the above-mentioned 3013 "diagnosing the target store and determining the target problem based on the diagnosis results" may include:

Diagnose the gap between the target store and the reference store, and/or the data differences of the target store at different times, and/or the gap between the store data of the target store and the set indicators, and obtain the diagnostic results;

The target problem is determined based on the diagnostic results.

The reference store can be: a store belonging to the same industry as the target store, whose data is at or above the median of all stores in that industry. Alternatively, the reference store can be: a store belonging to the same industry and level as the target store, whose data is at or above the median of all stores in that industry. Alternatively, the reference store can be: a store belonging to the same group and level as the target store, whose data is at or above the median of all stores in that industry. Alternatively, the reference store can be: a store specified by the user, and so on.

The aforementioned "data differences between different periods for the target store" can refer to: data differences between the target store's data in the first time period and the store's data in the second time period. The second time period may not overlap with the first time period, and the second time period may be earlier than the first time period. Alternatively, the second time period may partially overlap with the first time period, with the start time of the second time period earlier than the start time of the first time period, and the end time of the second time period earlier than the end time of the first time period. Or, the second time period may contain the first time period. That is, the start time of the first time period is a certain time within the second time period, and the end time of the first time period and the end time of the second time period are the same time.

The aforementioned "set metrics" can be user-defined target metrics. For example, if a user expects future store traffic to increase by 10%, this 10% increase is the set metric. Alternatively, the aforementioned "set metrics" can also be metrics that are used as a reference for other stores. See above for reference stores.

Metrics can include store metrics and product metrics. Store metrics include, but are not limited to: store page views, conversion rate, fan activity, member activity, average order value (the average amount spent by each customer who purchases a product), etc. Conversion rates include: search conversion rate, unique visitor (UV) order conversion rate, etc. Product metrics may include, but are not limited to: product views, product add-to-cart rate, product conversion rate, etc.

As can be seen from the above, the document database stores at least one factor of the query template obtained by attribution analysis of the preset question associated with the query template using a causal inference algorithm. Accordingly, step 302, "retrieving at least one factor of the target question from the document database," may include the following steps:

3021. Based on the target problem, determine the target query;

3022. In the document database, retrieve query templates that match the target query, and obtain at least one factor of the matched query template;

Specifically, when generating the Prompt (i.e., step 303 above), suggested content is generated based on at least one factor associated with the target Query and the matched Query template.

As mentioned above, a query template library has been pre-built, which includes multiple query templates. Accordingly, one possible implementation of step 3021, "determine the target query based on the target problem," is:

Based on the target problem, sort the multiple query templates in the query template library;

The target query is determined based on the first ranked query template, or the first number of ranked query templates are provided as candidate queries for the user to choose from, and the target query is determined based on the candidate query selected by the user.

In practice, the first number of items in the order can be 2, 3, or 5, etc. This embodiment does not make a specific limitation on this.

Furthermore, the copywriting database stores: store optimization items that can solve or improve the preset problem, obtained by attribution analysis of the preset problem associated with the query template using a causal inference algorithm, and at least one factor of the store optimization item. Correspondingly, step 3022 above, "retrieving query templates matching the target query from the copywriting database and obtaining at least one factor of the matched query templates," may include:

Retrieve a query template that matches the target query from the document database;

Obtain the store optimization items associated with the matched query template, and at least one factor of the store optimization items;

Specifically, when generating the Prompt (i.e., step 303 above), the suggested content is generated based on the target Query, the store optimization items associated with the matched Query template, and at least one factor associated with the store optimization items.

When constructing the copywriting database, this disclosure employs a causal inference algorithm. Specifically, the solution provided in this disclosure includes the following steps when constructing the copywriting database: using a causal inference algorithm to perform attribution analysis on a preset problem associated with a query template, obtaining store optimization items that solve or improve the preset problem, and at least one factor of the store optimization items. This part is an important aspect of the solution in this disclosure. One implementable method is that the step "using a causal inference algorithm to perform attribution analysis on a preset problem associated with a query template, obtaining store optimization items that solve or improve the preset problem, and at least one factor of the store optimization items" may include the following steps:

305. For the preset problem associated with the query template, funnel analysis technology is used to analyze the traffic data and operation data of the stores on the platform to identify store optimization items to solve or improve the preset problem;

306. Model the store optimization items as targets, and construct a causal tree model based on multiple factors;

307. Construct a dataset based on the traffic and operational data of stores on the platform;

308. Using the dataset, calculate the influence of each of the multiple factors on the prediction results of the causal tree model, and obtain the influence degree of each of the multiple factors;

309. Based on the influence of each of the multiple factors, sort the multiple factors to obtain the factor ranking of the store optimization items;

Wherein, at least one factor of the store optimization item is: the second-to-last factor in the factor ranking of the store optimization item.

For details on the implementation of steps 305 to 309 above, please refer to the description above, which will not be repeated here.

Furthermore, considering the differences between stores in different industries and at different levels, the factor ranking (factor priority) corresponding to the same optimization item for a store is different; the method provided in this embodiment also groups the stores on the platform to determine the factor ranking of store optimization items for different groups. That is, the method provided in this disclosure embodiment may further include the following steps:

310. Group the stores on the platform;

311. When constructing the dataset, it is built based on the traffic and operational data of stores within a group; and

312. Using the factor ranking of the store optimization items obtained from the dataset, the factor ranking of the store optimization items corresponding to the group is obtained.

Specifically, in the step of "obtaining the store optimization items associated with the matched query template and at least one factor of the store optimization items" (i.e., a subordinate step included in step 3022), the group to which the target store belongs is determined; the store optimization items associated with the matched query template and at least one factor of the store optimization items corresponding to the group to which the target store belongs are obtained.

The method provided in this embodiment is written from the perspective of how to apply the support provided by the technical architecture shown in Figure 1 to help users solve or improve store problems. In fact, the method provided in this embodiment also includes the following steps, which are part of the preliminary deployment tasks. Specifically, the method provided in this embodiment may further include the following steps:

313. Construct a query template library, which includes multiple query templates;

314. Based on the Query template library, construct the copywriting database;

315. Generate the training sample set based on the Query template library and the copywriting database;

316. Using the training sample set, adjust and train the pre-trained model.

For details on the implementation of steps 313 to 316 above, please refer to the description above, which will not be repeated here.

In one feasible embodiment, the Query template library includes at least one of the following: Query templates corresponding to the reasons for data differences between any store on the platform and a reference store, Query templates corresponding to the reasons for data differences between any store on the platform at different times, Query templates corresponding to store management issues, Query templates corresponding to store optimization items, and Query templates extended using a second preset model.

The copywriting database stores: attribute information of stores on the platform, traffic data and operational data of stores on the platform, store optimization items associated with query templates, factor sorting of store optimization items, suggested content associated with factors, and a corpus knowledge base of copywriting template information corresponding to different query templates.

The technical solution provided in this embodiment combines a large model with causal inference technology. Specifically, a causal inference algorithm is used when constructing the copywriting database to perform attribution analysis on preset questions associated with query templates to obtain at least one factor. The training samples used to adjust and train the pre-trained model are generated based on the query template library and the copywriting database. Using a training sample set composed of multiple such training samples, the pre-trained model is adjusted and trained to become a first preset model suitable for store operation scenarios. The copywriting database provides data support, retrieving at least one factor of store diagnostic results and/or user-inquired questions, as well as suggested content associated with that factor. The adjusted and trained first preset model provides intelligent response support. Based on store diagnostic results and/or user-inquired questions and suggested content associated with at least one factor, the first preset model can output copywriting that is more consistent with natural language logic and suitable for store operation scenarios, thus assisting users in operating their stores.

Another embodiment of this disclosure also provides an information processing method, more specifically a model training method. The executing entity of the method provided in this embodiment can be a server in the aforementioned service system. Specifically, as shown in Figure 4, the method includes:

401. Construct a query template library, which includes multiple query templates, and each query template is associated with a preset question related to store operations.

402. Based on the Query template library, construct a copywriting database; wherein the copywriting database stores at least one factor of the Query template obtained by attribution analysis of preset questions associated with the Query template using a causal inference algorithm.

403. Generate a training sample set based on the Query template library and the copywriting database.

404. Using the training sample set, the pre-trained model is adjusted and trained to obtain a first preset model that can handle intelligent response tasks in store operation scenarios.

When constructing the document database in step 402 above, attribution analysis of the preset questions associated with the query template is performed using a causal inference algorithm to obtain at least one factor of the query template, which may include:

4021. For the preset problem of Query template association, funnel analysis technology is used to analyze the traffic data and operation data of stores on the platform to identify store optimization items to solve or improve the preset problem;

4022. Model the store optimization items as targets, and construct a causal tree model based on multiple factors;

4023. Based on the traffic and operational data of stores on the platform, construct a dataset;

4024. Using the dataset, calculate the influence of each of the multiple factors on the prediction results of the causal tree model, and obtain the influence degree of each of the multiple factors;

4025. Based on the influence of each of the multiple factors, sort the multiple factors to obtain the factor ranking of the store optimization items.

Furthermore, the method provided in this embodiment may also include the following steps:

405. Group the stores on the platform.

406. When constructing the dataset, it is based on the traffic data and operational data of stores within a group.

407. Use the factor ranking of the store optimization items obtained from the dataset to rank the factors of the store optimization items corresponding to this group.

In one specific embodiment, step 401, "constructing a document database based on the Query template library," may include:

Acquire and store the attribute information of stores on the platform;

Acquire and store traffic and operational data of stores on the platform;

Attribution analysis of preset questions associated with query templates in the query template library is performed using a causal inference algorithm to obtain at least one factor of the query template;

Identify and store the suggested content for factor association;

Construct a corpus knowledge base of copywriting template information corresponding to different query templates.

In one specific embodiment, step 403 above, "generating a training sample set based on the Query template library and the copywriting database," may include:

The third preset model is used to output multiple training samples based on the Query template library and the copywriting database;

The training sample set includes the plurality of training samples.

Step 401, "Building a query template library," may include:

4011. Determine multiple basic query templates;

4012. Using the second preset model, the multiple basic query templates are extended to obtain multiple extended query templates;

4013. Evaluate the multiple extended query templates and retain the extended query templates that pass the evaluation;

The Query template library includes multiple basic Query templates and multiple extended Query templates that have passed evaluation.

For details of each step in this embodiment, please refer to the description above, which will not be repeated here.

This disclosure also provides a store diagnosis method according to an embodiment. The execution entity of the method provided in this embodiment can be a client in the aforementioned service system. As shown in Figure 5, the method includes:

501. Respond to diagnostic trigger events for the target store, trigger store diagnostics to present and/or play the diagnosed target issues;

502. Display and/or play text, said text including suggestions to help users solve or improve the target problem;

The text is obtained using a first preset model, which is obtained by adjusting and training the pre-trained model based on a pre-trained model for intelligent response tasks in a store operation scenario; the suggested content is associated with at least one factor of the target problem, and the at least one factor is obtained by attribution analysis of the target problem using a causal inference algorithm.

Referring to Figures 6a-6c, the diagnostic trigger events for the target store mentioned above may include:

Responding to user actions on diagnostic controls on the interface, generating diagnostic trigger events; and/or

Responding to the user's store diagnostic voice prompts, generating diagnostic trigger events; and/or

The system responds to store diagnostic commands input by users through the smart assistant interface, generating diagnostic trigger events.

Referring to the interface example shown in Figure 6a, this interface can be a store management interface, which displays diagnostic controls, such as "Store Diagnosis" shown in Figure 6a. Users can trigger the diagnosis by clicking the "Store Diagnosis" control.

Referring to the interface example shown in Figure 6b, the user can first wake up the smart assistant via voice (such as "Hello, smart assistant"), at which point the smart assistant icon 1 can be displayed on the interface. Then, the user can issue the voice command "Store Diagnosis" to trigger the diagnosis. The smart assistant icon can be any shape or pattern, and this embodiment does not limit it.

Referring to the interface example shown in Figure 6c, the display screen shows a smart assistant interaction interface 2. This smart interaction interface 2 displays an input box 3, where the user can enter "store diagnosis" and then click confirm or send control 4 (as shown in Figure 6c). This will trigger store diagnostics.

The aforementioned "store diagnosis" can be performed by the client. After triggering the diagnosis, the client can obtain store data of the reference store and/or historical store data of the target store from the server, and then call the client's local diagnostic program to perform a store diagnosis on the target store. Alternatively, the "store diagnosis" can be performed by the server. After the client triggers the diagnosis, it sends a target store diagnosis request to the server so that the server can obtain the target store's current store data, historical store data, reference store data, etc., and perform a diagnosis on the target store based on the obtained data to identify the target problem. The server then returns the identified target store to the client.

Specifically, "diagnosing the target store" may include: diagnosing the gap between the target store and the reference store, and/or the data differences of the target store at different times, and/or the gap between the store data of the target store and the set indicators, and obtaining diagnostic results.

The text displayed and/or played in the aforementioned 502 error can be provided by the server. For details, please refer to the description above, which will not be repeated here.

This disclosure also provides an embodiment of a method for operating a smart store assistant. The executing entity of the method provided in this embodiment can be a client in the aforementioned service system. As shown in Figure 7, the method includes:

601. Display the management page of the target store;

602. Monitor user actions on the management page to identify user-defined issues related to store operations;

603. In response to a user's confirmation command triggered for the target problem, display and/or play text, the text including suggestions to help the user solve or improve the target problem;

The text is obtained using a first preset model, which is obtained by adjusting and training the pre-trained model based on a pre-trained model for intelligent response tasks in a store operation scenario; the suggested content is associated with at least one factor of the target problem, and the at least one factor is obtained by attribution analysis of the target problem using a causal inference algorithm.

In the aforementioned 602, monitoring user actions on the management page to capture user-defined issues related to store operations may include:

6021. Respond to user input on the management page and determine the target problem based on the user input information; and/or

6022. Responding to a user's touch operation on a control on a management page, determining the target problem (e.g., how to set this setting) based on the settings associated with the control; and/or

6023. Detect a mouse cursor hovering over a content item on the management page, and based on that content item, determine the target problem; and/or

6024. Listen for store data viewing operations and determine the target problem based on the store data.

In the aforementioned 6021, users can directly input problem information in the smart assistant interaction area on the management page.

In the aforementioned 6022, for example, the setting could be a keyword setting. Users can set keywords for products in the keyword setting section. As shown in the figure, a control associated with the keyword setting is placed near it on the management page. When a user touches this control, the target question is determined to be "How to set keywords?"

In case 6023 above, if a user lingers on a particular content item for an extended period, there is a higher probability that they have questions. Therefore, if the mouse cursor is detected to linger on a content item for a certain duration (e.g., 10 seconds, 30 seconds, etc.), the target question can be determined based on that content item.

In the case of 6024 above, if a user views the store's recent traffic through the management page, the target question can be determined based on the store traffic the user is currently seeing, such as "Why is the traffic decreasing?".

Furthermore, the method provided in this embodiment may also include the following steps:

604. Display the smart assistant icon on the management page corresponding to the target store;

605. In response to the smart assistant wake-up operation, present a smart assistant page containing an input box and at least one recommended query; wherein, at least one recommended query is obtained by diagnosing the target store.

Users can enter their target problem in the input box and interact with the intelligent assistant to obtain the necessary data, solutions, and more.

The methods provided in the two embodiments above are written from the perspective of an intelligent assistant, corresponding to the product form of the technical solutions provided in this disclosure. Based on the technical support provided by the aforementioned server, this intelligent assistant can help merchants improve analysis efficiency, provide effective strategies, and seize market opportunities, thus helping merchants grow better. The intelligent assistant can diagnose store problems, use causal inference technology to analyze and identify store optimization items to solve or improve problems, as well as at least one factor of the store optimization items. Then, using a large model, it outputs response text that conforms to human language logic and fits the store's operational scenario, thereby assisting merchants in their growth.

This disclosure also provides an information processing apparatus in one embodiment. The apparatus includes a determining module, a retrieving module, a generating module, and a calculating module. The determining module identifies a target problem for a target store. The retrieving module retrieves at least one factor of the target problem from a document database. The generating module generates a prompt based on suggested content associated with the target problem and the at least one factor. The calculating module outputs text corresponding to the prompt using a first preset model to help the user solve or improve the target problem.

The document database stores at least one factor of the preset question obtained by attribution analysis of the preset question using a causal inference algorithm; the first preset model is obtained by adjusting and training the pre-trained model based on the pre-trained model for intelligent response tasks in the store operation scenario.

Furthermore, the document database stores at least one factor of the query template obtained by attribution analysis of a preset question associated with the query template using a causal inference algorithm. The retrieval module is used to determine the target query based on the target question; retrieve query templates matching the target query from the document database, and obtain at least one factor of the matched query templates; wherein, when generating the prompt, suggested content is generated based on the target query and at least one factor associated with the matched query template.

Furthermore, when determining the target query based on the target question, the determining module sorts multiple query templates in the query template library according to the target question; determines the target query based on the first ranked query template, or provides the first number of ranked query templates as candidate queries for the user to choose from, and determines the target query based on a candidate query selected by the user.

Furthermore, the copywriting database stores: store optimization items that can solve or improve the preset problems associated with the query template, obtained by attribution analysis using a causal inference algorithm, and at least one factor of the store optimization items. When the retrieval module retrieves query templates matching the target query from the copywriting database and obtains at least one factor of the matched query templates, it specifically performs the following: retrieving query templates matching the target query from the copywriting database; obtaining store optimization items associated with the matched query templates, and at least one factor of the store optimization items; wherein, when generating the Prompt, suggested content is generated based on the target query, the store optimization items associated with the matched query templates, and at least one factor of the store optimization items.

Furthermore, the apparatus provided in this embodiment may also include a copywriting database construction module. This copywriting database construction module can utilize a causal inference algorithm to perform attribution analysis on preset questions associated with the query template, obtaining store optimization items that solve or improve the preset questions, and at least one factor of the store optimization items. Specifically, the copywriting database construction module may be used for:

For the preset problem associated with the query template, funnel analysis technology is used to analyze the traffic and operational data of stores on the platform to identify store optimization items that solve or improve the preset problem. A causal tree model is constructed based on multiple factors, with the store optimization items as the target. A dataset is built based on the traffic and operational data of stores on the platform. Using the dataset, the influence of each factor on the prediction results of the causal tree model is calculated to obtain the influence degree of each factor. The factors are ranked according to their respective influence degrees to obtain the factor ranking of the store optimization items. At least one factor of each store optimization item is the second-to-last ranked factor in the factor ranking of the store optimization items.

Furthermore, the document database construction module in the device provided in this embodiment is also used for: grouping stores on the platform; constructing the dataset based on traffic data and operational data of stores within a group; and using the factor ranking corresponding to the store optimization items obtained from the dataset to perform factor ranking of the store optimization items corresponding to the group. Specifically, when obtaining the store optimization items associated with the matched query template and at least one factor of the store optimization items, the group to which the target store belongs is determined; the store optimization items associated with the matched query template and at least one factor of the store optimization items corresponding to the group to which the target store belongs are obtained.

Furthermore, the apparatus provided in this embodiment also includes a Query template library construction module, a sample generation module, and an adjustment and training module. The Query template library construction module is used to construct a Query template library, which includes multiple Query templates. The document data construction module is used to construct a document data library based on the Query template library. The sample generation module is used to generate the training sample set based on the Query template library and the document data library. The adjustment and training module is used to adjust and train the pre-trained model using the training sample set.

Furthermore, when determining the target problem for a target store, the determining module is specifically used for:

Responding to user input, determining the target question for the target store based on the user input information; and/or

Responding to a user request triggered on the management page of a target store regarding a setting item, and for that setting item, determining the target issue; and/or

The target store is diagnosed, and the target problem is determined based on the diagnosis results; and/or

Obtain at least one recommendation question from the network side based on big data analysis, and in response to a selection event, take the selected recommendation question as the target question.

Specifically, when the determination module diagnoses the target store and determines the target problem based on the diagnosis results, it is used to: diagnose the gap between the target store and the reference store, and/or the data differences of the target store at different times, and/or the gap between the store data of the target store and the set indicators, and obtain the diagnosis results; and determine the target problem based on the diagnosis results.

Another embodiment of this disclosure provides an information processing apparatus comprising: a query template library construction module, a copywriting database construction module, a sample generation module, and an adjustment and training module. The query template library construction module is used to construct a query template library, which includes multiple query templates, each associated with a preset question related to store operations. The copywriting database construction module is used to construct a copywriting database based on the query template library; wherein the copywriting database stores at least one factor of the query template obtained by attribution analysis of the preset question associated with the query template using a causal inference algorithm. The sample generation module is used to generate a training sample set based on the query template library and the copywriting database. The adjustment and training module is used to adjust and train a pre-trained model using the training sample set to obtain a first preset model capable of handling intelligent response tasks in store operation scenarios.

In one specific implementation scheme, when the above-mentioned copywriting database construction module constructs the copywriting database, it uses a causal inference algorithm to perform attribution analysis on the preset questions associated with the query template to obtain at least one factor of the query template. Specifically, this is used for:

For the preset problem associated with the query template, funnel analysis technology is used to analyze the traffic and operational data of stores on the platform to identify store optimization items that solve or improve the preset problem. A causal tree model is constructed based on multiple factors, with the store optimization items as the target model. A dataset is built based on the traffic and operational data of stores on the platform. Using the dataset, the influence of each factor on the prediction results of the causal tree model is calculated to obtain the influence degree of each factor. Based on the influence degree of each factor, the factors are ranked to obtain the factor ranking of the store optimization items.

Furthermore, the copywriting database construction module is also used to: group the stores on the platform; construct the dataset based on the traffic data and operational data of the stores within a group; and use the factor ranking of the store optimization items obtained from the dataset to perform factor ranking of the store optimization items corresponding to the group.

Furthermore, when constructing the copywriting database based on the Query template library, the copywriting database construction module is specifically used for: acquiring and storing attribute information of stores on the platform; acquiring and storing traffic data and operational data of stores on the platform; using a causal inference algorithm to perform attribution analysis on preset questions associated with Query templates in the Query template library to obtain at least one factor of the Query template; determining and storing suggested content associated with the factor; and constructing a corpus knowledge base of copywriting template information corresponding to different Query templates.

When generating a training sample set based on the Query template library and the copywriting database, the sample generation module is specifically used to: output multiple training samples based on the Query template library and the copywriting database using a third preset model; wherein, the training sample set includes the multiple training samples.

Furthermore, when constructing the query template library, the query template library construction module is specifically used to: determine multiple basic query templates; expand the multiple basic query templates using a second preset model to obtain multiple extended query templates; evaluate the multiple extended query templates and retain the extended query templates that pass the evaluation; wherein, the query template library includes multiple basic query templates and multiple extended query templates that have passed the evaluation.

It should be noted that the information processing devices provided in the above two embodiments can implement the technical solutions described in the corresponding method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content of each method embodiment, and will not be repeated here.

An embodiment of this disclosure provides a store diagnostic device including a trigger module and an output module. The trigger module responds to a diagnostic trigger event for a target store, triggering store diagnostics to present and/or play the diagnosed target problem. The output module displays and/or plays text, including suggestions to help users solve or improve the target problem. The text is obtained using a first preset model, which is trained and adjusted based on a pre-trained model for intelligent response tasks in a store operation scenario. The suggestions are associated with at least one factor of the target problem, and the at least one factor is determined by attribution analysis of the target problem using a causal inference algorithm.

It should be noted that the store diagnostic device provided in the above embodiments can realize the technical solutions described in the corresponding method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the corresponding method embodiments, and will not be repeated here.

An embodiment of this disclosure provides a smart store assistant device including a display device, a monitoring and capture module, and an output module. The display module displays the management page of the target store. The monitoring and capture module monitors user operations on the management page to capture target issues related to store operations. The output module responds to confirmation commands triggered by the user regarding the target issues, displaying and/or playing text, including suggestions to help the user resolve or improve the target issues.

The text is obtained using a first preset model, which is obtained by adjusting and training the pre-trained model based on a pre-trained model for intelligent response tasks in a store operation scenario; the suggested content is associated with at least one factor of the target problem, and the at least one factor is obtained by attribution analysis of the target problem using a causal inference algorithm.

It should be noted that the smart assistant device for stores provided in the above embodiments can implement the technical solutions described in the corresponding method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above method embodiments, and will not be repeated here.

An embodiment of this disclosure also provides an electronic device. As shown in FIG8, the electronic device includes a processor 42 and a memory 41. The memory 41 is used to store one or more computer programs; the processor 42 is coupled to the memory 41 and is used for the at least one or more computer programs to implement the steps in the methods provided in the embodiments of this disclosure.

The aforementioned memory 41 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

Furthermore, the electronic device also includes other components such as a communication component 43, a display 44, a power supply component 45, and an audio component 46. Only some components are shown here for illustrative purposes, and it is not intended that the electronic device contains only these components.

Accordingly, this disclosure also provides a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the method steps or functions provided in the embodiments of this disclosure described above.

The methods in this disclosure can be implemented, in whole or in part, by software, hardware, firmware, or any combination thereof. When implemented in software, they can be implemented, in whole or in part, as a computer program product. Therefore, this disclosure also provides a computer program product. This computer program product includes a computer program/instructions that, when executed by an electronic component such as a processor, can perform, in whole or in part, the steps or functions of the methods provided in the embodiments of this disclosure. The computer can be a general-purpose computer, a special-purpose computer, a computer network, network equipment, user equipment, core network equipment, OAM, or other programmable device.

Claims (22)

An information processing method, comprising: For the target store, identify the target problem; Retrieve at least one factor of the target question from the document database; Based on the target question and the suggested content associated with at least one factor, a prompt word is generated. The first preset model is used to output the text corresponding to the Prompt, so as to help users solve or improve the target problem; The document database stores at least one factor of the preset question obtained by attribution analysis of the preset question using a causal inference algorithm; the first preset model is obtained by adjusting and training the pre-trained model based on the pre-trained model for intelligent response tasks in the store operation scenario. According to the method of claim 1, wherein the document database stores: at least one factor of the query template obtained by attribution analysis of preset questions associated with the query template using a causal inference algorithm; and Retrieve at least one factor of the target question from the document database, including: Based on the target question, determine the target query; In the document database, retrieve query templates that match the target query and obtain at least one factor of the matched query template; Specifically, when generating the Prompt, suggested content is generated based on at least one factor associated with the target Query and the matched Query template. According to the method of claim 2, determining the target query based on the target question includes: Based on the target problem, sort the multiple query templates in the query template library; The target query is determined based on the first ranked query template, or the first number of ranked query templates are provided as candidate queries for the user to choose from, and the target query is determined based on the candidate query selected by the user. According to the method of claim 2 or 3, wherein the copywriting database stores: store optimization items that can solve or improve the preset problems, and at least one factor of the store optimization items, obtained by attribution analysis of preset problems associated with query templates using a causal inference algorithm; and In the document database, retrieve query templates that match the target query, and obtain at least one factor of the matched query templates, including: Retrieve a query template that matches the target query from the document database; Obtain the store optimization items associated with the matched query template, and at least one factor of the store optimization items; Specifically, when generating the Prompt, the suggested content is generated based on the target Query, the store optimization items associated with the matched Query template, and at least one factor associated with the store optimization items. According to the method of claim 4, the attribution analysis of the preset problem associated with the query template using a causal inference algorithm to obtain store optimization items that solve or improve the preset problem and at least one factor of the store optimization items includes: To address the pre-defined issues associated with query templates, funnel analysis techniques are used to analyze the traffic and operational data of stores on the platform in order to identify store optimization items that can solve or improve the pre-defined issues. Modeling is performed using the aforementioned store optimization items as targets, and a causal tree model is constructed based on multiple factors; A dataset is constructed based on the traffic and operational data of stores on the platform; Using the dataset, the influence of each of the multiple factors on the prediction results of the causal tree model is calculated, and the influence degree of each of the multiple factors is obtained. Based on the influence of each of the multiple factors, the multiple factors are sorted to obtain the factor ranking of the store optimization items; Wherein, at least one factor of the store optimization item is: the second-to-last factor in the factor ranking of the store optimization item. The method according to claim 5, further comprising: Group the stores on the platform; The dataset was constructed based on traffic and operational data of stores within a group; and The factor ranking corresponding to the store optimization item obtained using the dataset is the factor ranking of the store optimization item corresponding to this group. Specifically, when obtaining the store optimization items associated with the matched query template and at least one factor of the store optimization items, the group to which the target store belongs is determined; the store optimization items associated with the matched query template and at least one factor of the store optimization items corresponding to the group to which the target store belongs are obtained. The method according to any one of claims 1 to 6, further comprising: Build a query template library, which includes multiple query templates; Based on the Query template library, the copywriting database is constructed; The training sample set is generated based on the Query template library and the copywriting database; The pre-trained model is adjusted and trained using the training sample set. According to the method of claim 7, the Query template library comprises at least one of the following: The query template for the reason why the data of any store on the platform differs from that of a reference store; Query templates for explaining why data discrepancies exist for any store on the platform at different times; Query templates for store management issues; Query templates corresponding to store optimization items; The query template is extended using the second preset model; The document database stores: The attribute information of stores on the platform; Traffic and operational data of stores on the platform; Store optimization items associated with the query template; Factor sorting of store optimization items; Suggested content regarding factor correlation; A corpus knowledge base of copywriting template information corresponding to different query templates. According to the method of claim 8, wherein determining the target problem for the target store includes: Responding to user input, determining the target question for the target store based on the user input information; and/or Responding to a user request triggered on the management page of a target store regarding a setting item, and for that setting item, determining the target issue; and/or The target store is diagnosed, and the target problem is determined based on the diagnosis results; and/or Obtain at least one recommendation question from the network side based on big data analysis, and in response to a selection event, take the selected recommendation question as the target question. According to the method of claim 9, wherein diagnosing the target store and determining the target problem based on the diagnosis results includes: Diagnose the gap between the target store and the reference store, and/or the data differences of the target store at different times, and/or the gap between the store data of the target store and the set indicators, and obtain the diagnostic results; The target problem is determined based on the diagnostic results. An information processing method, comprising: Construct a query template library, which includes multiple query templates, and each query template is associated with a preset question related to store operations; Based on the Query template library, a copywriting database is constructed; wherein, the copywriting database stores at least one factor of the Query template obtained by attribution analysis of preset questions associated with the Query template using a causal inference algorithm; A training sample set is generated based on the Query template library and the copywriting database; Using the training sample set, the pre-trained model is adjusted and trained to obtain a first preset model that can handle intelligent response tasks in store operation scenarios. According to the method of claim 11, wherein, when constructing the document database, attribution analysis is performed on the preset questions associated with the query template using a causal inference algorithm to obtain at least one factor of the query template, including: To address the pre-defined issues associated with query templates, funnel analysis techniques are used to analyze the traffic and operational data of stores on the platform in order to identify store optimization items that can solve or improve the pre-defined issues. Modeling is performed using the aforementioned store optimization items as targets, and a causal tree model is constructed based on multiple factors; A dataset is constructed based on the traffic and operational data of stores on the platform; Using the dataset, the influence of each of the multiple factors on the prediction results of the causal tree model is calculated, and the influence degree of each of the multiple factors is obtained. Based on the influence of each of the multiple factors, the factors are sorted to obtain the factor ranking of the store optimization items. The method according to claim 12, further comprising: Group the stores on the platform; The dataset was constructed based on traffic and operational data of stores within a group; and The factor ranking of the store optimization items obtained using the dataset is used to rank the store optimization items corresponding to this group. The method according to any one of claims 11 to 13, wherein constructing a document database based on the Query template library includes: Acquire and store the attribute information of stores on the platform; Acquire and store traffic and operational data of stores on the platform; Attribution analysis of preset questions associated with query templates in the query template library is performed using a causal inference algorithm to obtain at least one factor of the query template; Identify and store the suggested content for factor association; Construct a corpus knowledge base of copywriting template information corresponding to different query templates. The method according to any one of claims 11 to 14, wherein generating a training sample set based on the Query template library and the copywriting database includes: The third preset model is used to output multiple training samples based on the Query template library and the copywriting database; The training sample set includes the plurality of training samples. The method according to any one of claims 11 to 15, wherein constructing the query template library includes: Identify multiple basic query templates; Using the second preset model, the multiple basic query templates are expanded to obtain multiple extended query templates; The multiple extended query templates are evaluated, and the extended query templates that pass the evaluation are retained; The Query template library includes multiple basic Query templates and multiple extended Query templates that have passed evaluation. A store diagnostic method, applicable to the client, the method includes: Respond to diagnostic trigger events for the target store, trigger store diagnostics to present and/or play the diagnosed target issues; Display and/or play text, which includes suggestions to help users solve or improve the target problem; The text is obtained using a first preset model, which is obtained by adjusting and training the pre-trained model based on a pre-trained model for intelligent response tasks in a store operation scenario; the suggested content is associated with at least one factor of the target problem, and the at least one factor is obtained by attribution analysis of the target problem using a causal inference algorithm. A method for operating a smart store assistant, applicable to the client side, the method includes: Display the target store's management page; Monitor user actions on the management page to identify user-defined issues related to store operations; In response to a user's confirmation command triggered for the target problem, display and/or play text, which includes suggestions to help the user solve or improve the target problem; The text is obtained using a first preset model, which is obtained by adjusting and training the pre-trained model based on a pre-trained model for intelligent response tasks in a store operation scenario; the suggested content is associated with at least one factor of the target problem, and the at least one factor is obtained by attribution analysis of the target problem using a causal inference algorithm. A service system, comprising: A client for implementing the method described in claim 17 or 18 above; A server-side component for implementing the method described in any one of claims 1 to 16. An electronic device, comprising a memory and a processor; wherein, The memory is used to store executable instructions; The processor executes the executable instructions to implement the steps in the information processing method as described in any one of claims 1 to 16, the steps in the store diagnosis method as described in claim 17, or the steps in the store smart assistant working method as described in claim 18. A computer-readable storage medium storing computer instructions that, when executed by a processor, can perform the steps of the information processing method as described in any one of claims 1 to 16, the steps of the store diagnosis method as described in claim 17, or the steps of the store smart assistant working method as described in claim 18. A computer program product comprising a computer program or instructions that, when executed by a processor, cause the processor to perform steps in the information processing method of any one of claims 1 to 16, or steps in the store diagnosis method of claim 17, or steps in the store smart assistant working method of claim 18.