Form configuration method and apparatus, computer-readable storage medium, and product

Abstract

The present disclosure relates to a form configuration method and device, a computer readable storage medium and product, and relates to the technical field of information processing. The form configuration method comprises: obtaining configured content in a form, wherein the configured content comprises a title and a first field; and generating a recommended configuration field of the form, wherein the recommended configuration field is determined from a plurality of candidate fields according to semantic information of the title and semantic information of the first field after determining an overall intention of the form and a local intention of the first field.

Description

Technical Field

[0001] This disclosure relates to the field of information processing technology, and in particular to a form configuration method, apparatus, computer-readable storage medium, and product. Background Technology

[0002] Using forms to collect data is a common information processing method. Those being collected fill out the forms according to the titles and fields specified, and the data collector then compiles the data by integrating the completed forms. Summary of the Invention

[0003] One of the technical problems this disclosure aims to solve is: how to improve the efficiency of form configuration.

[0004] According to a first aspect of some embodiments of this disclosure, a form configuration method is provided, comprising: obtaining configured content in a form, wherein the configured content includes a title and a first field; generating recommended configuration fields for the form, wherein the recommended configuration fields are determined from a plurality of candidate fields after determining the overall intent of the form and the partial intent of the first field based on the semantic information of the title and the semantic information of the first field.

[0005] In some embodiments, the configured content includes multiple first fields. Determining the local intent of the first field based on the semantic information of the title and the semantic information of the first field includes: determining the associated field of each first field based on the multiple first fields; and determining the local intent of the first field by fusing the semantic information of the first field and the semantic information of the associated field of the first field.

[0006] In some embodiments, determining the associated field of each first field based on a plurality of first fields includes: determining the association relationship between the plurality of first fields based on entities and relationships between entities in the city knowledge graph; and for each of the plurality of first fields, determining the associated field of the first field from other first fields besides the first field in the plurality of first fields based on the association relationship between the plurality of first fields.

[0007] In some embodiments, the configured content includes multiple first fields, and the form configuration method further includes: for each of the multiple first fields, using at least one of a city knowledge graph and a large language model to perform semantic enhancement on the first field to obtain semantic enhancement information of the first field; determining the association relationship of the multiple first fields based on the city knowledge graph; and for each first field, processing the semantic enhancement information of the first field using the association relationship of the multiple first fields to determine the semantic information of the first field.

[0008] In some embodiments, using at least one of a city knowledge graph and a large language model to semantically enhance the first field to obtain semantically enhanced information of the first field includes: using a city knowledge graph to perform a first semantic enhancement on the first field to obtain first semantically enhanced information; using a large language model to perform a second semantic enhancement on the first field to obtain second semantically enhanced information; and combining at least one of the first semantically enhanced information and the second semantically enhanced information with the first field to obtain semantically enhanced information of the first field.

[0009] In some embodiments, the semantic enhancement information of the first field is processed by utilizing the association relationship of multiple first fields to determine the semantic information of the first field, which includes: obtaining a first vector of the first field by fusing at least one of the first semantic enhancement information and the second semantic enhancement information and the first field, wherein the first vector represents the semantic enhancement information of the first field; and processing the first vector of each first field and the association relationship of multiple first fields by a graph neural network model to obtain a second vector, wherein the second vector represents the semantic information of the first field.

[0010] In some embodiments, the configured content includes multiple first fields. Determining the overall intent of the form based on the semantic information of the title and the semantic information of the first fields includes: for each of the multiple first fields, generating overall semantic information of the first field by combining the semantic information of the title and the semantic information of the first field; processing the overall semantic information of each first field using a graph neural network model; and determining the overall intent of the form based on the processed overall semantic information of the multiple first fields.

[0011] In some embodiments, the overall intent of the form is determined using a first model trained with first positive samples and first negative samples, wherein the first positive samples are generated based on sub-forms obtained from the same form and the first negative samples are generated based on sub-forms obtained from different forms; the local intent of the first field is determined using a second model trained with second positive samples and second negative samples, wherein the second positive samples are generated based on fields obtained from the same form and the second negative samples are generated based on fields obtained from different forms.

[0012] In some embodiments, the form configuration method further includes: determining a target intent based on the overall intent of the form and the local intent of the first field; determining a recommended value for each candidate field based on the similarity between the target intent and each candidate field among a plurality of candidate fields; and determining a recommended configuration field based on the recommended value of each recommended field.

[0013] In some embodiments, determining the target intent based on the overall intent of the form and the partial intent of the first field includes: determining an intermediate intent by fusing the overall intent of the form and the partial intent of the first field; generating target parameters based on the overall intent of the form; and processing the intermediate intent based on the target parameters to generate the target intent.

[0014] In some embodiments, generating the target parameters based on the overall intent of the form includes: generating a first matrix parameter and a second matrix parameter based on the overall intent of the form, wherein the rank of the first matrix parameter and the second matrix parameter is lower than a specified value; and generating the target parameters based on the first matrix parameter and the second matrix parameter.

[0015] In some embodiments, the form configuration method further includes: in response to a recommended configuration field being configured in the form, retrieving the configured content in the form to generate the next recommended configuration field for the form, until the form configuration is complete.

[0016] According to a second aspect of some embodiments of the present disclosure, a form configuration apparatus is provided, comprising: an acquisition module configured to acquire configured content in a form, wherein the configured content includes a title and a first field; and a generation module configured to generate recommended configuration fields for the form, wherein the recommended configuration fields are determined from a plurality of candidate fields after determining the overall intent of the form and the partial intent of the first field based on the semantic information of the title and the semantic information of the first field.

[0017] According to a third aspect of some embodiments of the present disclosure, a form configuration apparatus is provided, including: a processor; and a memory coupled to the processor for storing instructions, which, when executed by the processor, cause the processor to perform the form configuration method as described above.

[0018] According to a fourth aspect of some embodiments of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon, wherein the program, when executed by a processor, implements the form configuration method as described above.

[0019] According to a fifth aspect of some embodiments of the present disclosure, a computer program product is provided, including instructions that, when executed by a processor, cause the processor to perform the form configuration method as described above.

[0020] This disclosure analyzes the overall intent of the form and the partial intent of its fields based on the content already configured in the form, and then determines the recommended configuration field from multiple candidate fields. The form configuration method of this disclosure eliminates the need for manual full form configuration; it can utilize manually configured partial content to recommend configuration fields, thus effectively improving the efficiency of form configuration.

[0021] Other features and advantages of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0022] 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 only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating form configuration according to some embodiments of this disclosure is shown.

[0024] Figure 2 A flowchart illustrating the determination of semantic information of a first field according to some embodiments of the present disclosure is shown.

[0025] Figure 3 A flowchart illustrating a partial intent for determining a first field according to some embodiments of the present disclosure is shown.

[0026] Figure 4 A flowchart illustrating the overall intent of determining a form according to some embodiments of this disclosure is shown.

[0027] Figure 5 A flowchart illustrating the determination of recommended configuration fields according to some embodiments of this disclosure is shown.

[0028] Figure 6 A flowchart of a form configuration apparatus according to some embodiments of the present disclosure is shown.

[0029] Figure 7 A schematic diagram of the structure of a form configuration apparatus according to other embodiments of the present disclosure is shown.

[0030] Figure 8 A schematic diagram of the structure of a form configuration apparatus according to some embodiments of the present disclosure is shown. Detailed Implementation

[0031] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0032] Using forms for information statistics typically involves multiple personnel. For example, in information collection scenarios at the grassroots community level, each community collects information independently, resulting in different personnel configuring the forms for each community. Because these different personnel often configure the forms based on their own personal habits, data from different forms becomes difficult to align. For instance, for the same information, different personnel might use different fields to represent it in the forms.

[0033] To facilitate form alignment, a city knowledge graph (or city knowledge system) can be used to standardize the mapping of fields in the form. The city knowledge graph includes a series of standard terms covering various aspects of city knowledge, which can serve as a standard for form alignment. Thus, during form configuration, administrators can select the fields to be configured in the form based on the field descriptions of the current business scenario, and then achieve alignment between different forms based on the mapping relationship between the fields and the standard terms.

[0034] However, while the standard terminology in a city knowledge graph is limited, the space for corresponding field descriptions is vast. Furthermore, there is significant similarity and redundancy among field descriptions; for example, "temporary address" and "temporary residence address" have the same meaning, further increasing the space for field descriptions. If all field descriptions were displayed for configuration personnel to browse, the selection cost would be extremely high, easily creating reading difficulties, increasing configuration complexity, reducing configuration efficiency, and even preventing configuration from being completed.

[0035] Based on this, this disclosure provides a form configuration method.

[0036] Figure 1 A flowchart illustrating a form configuration method according to some embodiments of this disclosure is shown. Figure 1 As shown, the method of this embodiment includes steps S11 and S12.

[0037] In step S11, the configured content in the form is obtained, including the title and the first field.

[0038] The form title explains the overall purpose of the information collection. For example, if the form title is "Annual Grade Statistics for Class XX," it means the purpose of the form is to collect student grades. Therefore, the form should include fields for collecting grades.

[0039] Forms collect information through fields, with each field responsible for collecting a specific type of information. For example, a form titled "Annual Grade Statistics for Class XX" might include fields such as Chinese language score and math score. In other words, each field also has a specific purpose for information collection. Each field reflects the user's specific needs within the context of that field; for instance, in the "Home Address" field, a user might need to add "Postal Code" or "Street Address."

[0040] By retrieving the configured content from the form, including the title and one or more first fields, we can analyze the business logic that the configuration personnel expect to express, and then recommend configuration fields.

[0041] In step S12, recommended configuration fields for the form are generated. These recommended configuration fields are determined from multiple candidate fields based on the semantic information of the title and the semantic information of the first field, after determining the overall intent of the form and the partial intent of the first field.

[0042] The intent of a form comprises two dimensions: the overall intent of the form and the local intent of each field. Determining the overall intent of the form and the local intent of each first field based on the configured title and first field allows for a thorough understanding of the configuration user's business logic, thereby predicting the fields that need to be configured in the form and displaying them as recommended configuration fields to the configuration user. The configuration user can, for example, determine whether to configure the recommended configuration fields in the form by triggering a control. In other words, the apparatus for executing the form configuration method of this disclosure is user-interactive, for example, interacting with the user (i.e., the configuration user) through a form configuration application. On the form configuration application, the user configures the title and one or more first fields, and then the form configuration application displays the recommended configuration fields determined by executing the form configuration method of this disclosure to the user, who can then configure the form by selecting the recommended configuration fields.

[0043] Candidate fields here can be selected from historical fields based on the current business scenario. The current business scenario can be determined by the configuration personnel, or it can be automatically determined based on the content already configured in the form. For example, in a grade statistics scenario, fields used in grade statistics will be used as candidate fields. Alternatively, candidate fields can also be standard terms selected from the city knowledge graph based on the current business scenario, which facilitates quick alignment between different forms in the future.

[0044] This disclosure analyzes the overall intent of the form and the partial intent of its fields based on the content already configured in the form, and then determines the recommended configuration field from multiple candidate fields. The form configuration method of this disclosure eliminates the need for manual full form configuration; it can utilize manually configured partial content to recommend configuration fields, thus effectively improving the efficiency of form configuration.

[0045] The process of determining the recommended configuration fields is described in detail below.

[0046] As mentioned above, both the overall intent of the form and the specific intent of the first field require the use of its semantic information. To fully reflect the semantic information of the first field, semantic enhancement can be applied to it, reducing the problem of semantic information bias caused by insufficient semantic description of the first field itself.

[0047] In some embodiments, the configured content includes multiple first fields, and the form configuration method further includes: for each of the multiple first fields, using at least one of a city knowledge graph and a large language model to perform semantic enhancement on the first field to obtain semantic enhancement information of the first field; determining the association relationship of the multiple first fields based on the city knowledge graph; and for each first field, processing the semantic enhancement information of the first field using the association relationship of the multiple first fields to determine the semantic information of the first field.

[0048] Since the first field can be mapped to standard terms in the city knowledge graph, its semantics can be expanded using the structured information in the city knowledge graph. For example, information that can be added to "home address" includes geographical coordinates and administrative divisions.

[0049] Furthermore, since all these first fields can be mapped to a city knowledge graph, which categorizes its standard terms into five types of entities and describes the relationships between these entities, the associations of multiple first fields can be determined using the entities and relationships between them in the city knowledge graph. For example, a first heterogeneous graph of these multiple first fields can be constructed using the city knowledge graph. Nodes in the first heterogeneous graph represent first fields, and edges represent the associations between them.

[0050] Considering the limited information in the city knowledge graph, a large language model can be used to semantically expand the first field. For example, a detailed definition and usage context can be generated for the first field, thereby injecting richer semantic information. This can effectively reduce the problem of insufficient semantic information caused by the limited semantic information of the first fragment in the form and the lack of integration with external knowledge, thus improving the accuracy of inferring user intent and consequently improving recommendation accuracy.

[0051] In some embodiments, using at least one of a city knowledge graph and a large language model to semantically enhance the first field to obtain semantically enhanced information of the first field includes: using a city knowledge graph to perform a first semantic enhancement on the first field to obtain first semantically enhanced information; using a large language model to perform a second semantic enhancement on the first field to obtain second semantically enhanced information; and combining at least one of the first semantically enhanced information and the second semantically enhanced information with the first field to obtain semantically enhanced information of the first field.

[0052] After determining the semantic enhancement information of the first field, the semantic information of the first field is finally determined by combining the structural information of the form, i.e., the relationship between multiple first fields. For example, a graph neural network model can be used to process the semantic enhancement information of multiple first fields and the relationship between multiple fields to obtain a second heterogeneous graph. The nodes in the second heterogeneous graph represent the first field and its semantic information, and the edges in the second heterogeneous graph represent the structural and semantic relationships between the nodes.

[0053] In some embodiments, the semantic enhancement information of a first field is processed using the association relationships of multiple first fields to determine the semantic information of the first field. This process includes: fusing at least one of the first semantic enhancement information and the second semantic enhancement information with the first field to obtain a first vector of the first field, wherein the first vector represents the semantic enhancement information of the first field; and processing the first vector of each first field and the association relationships of multiple first fields using a graph neural network model to obtain a second vector, wherein the second vector represents the semantic information of the first field. The graph neural network model is, for example, a heterogeneous graph transformer.

[0054] By unifying semantic information from different sources into a single representation space—for example, concatenating the first semantic enhancement information, the second semantic enhancement information, and the first field—and encoding them using a pre-trained multilingual large language model (e.g., the XLM-RoBERTa model), an initial representation of the first field, i.e., the first vector, is obtained. Then, a graph neural network model (e.g., a heterogeneous graph transformer) is used to perform deep structure-aware information aggregation on the first vector of the first field and the first heterogeneous graph. After aggregation, a second heterogeneous graph is obtained, and the nodes in the second heterogeneous graph are represented by the second vector. The second vector is a hybrid representation of the information enhanced by the city knowledge graph, the semantic information enhanced by the large language model, and the form structure information. The second vector not only reflects the semantics of the first field itself but also the contextual relationship of the first field within the form, providing a richer feature foundation for subsequent intent learning and recommendation matching.

[0055] Figure 2 A flowchart illustrating the determination of semantic information of a first field according to some embodiments of the present disclosure is shown.

[0056] In step S21, the first field is semantically augmented using a Large Language Model (LLM) and an Urban Knowledge System (UKS) to obtain first semantic enhancement information and second semantic enhancement information.

[0057] In step S22, the first semantic enhancement information, the second semantic enhancement information, and the first field are input into the XLM-RoBERTa model (as an example of a multilingual large language model).

[0058] In step S23, the output of the XLM-RoBERTa model is processed using a heterogeneous graph transformer (as an example of a graph neural network model) to output the semantic information of the first field.

[0059] The above process describes the determination of the semantic information of the first field. In determining the semantic information of the first field, heterogeneous graph generation is utilized. By generating a heterogeneous graph, semantic information from different sources can be mixed to generate a vector representation of the semantic information of the first field, facilitating the subsequent determination of the overall and local intents.

[0060] Similar to the linguistic information in the first field, the semantic information in the title can also be semantically enhanced using urban knowledge graphs and large language models, which will not be elaborated here.

[0061] After determining the semantic information of the title and the first field, the overall intent of the form and the local intent of the first field are determined. These two intents are determined using separate representation learning mechanisms, that is, by using different methods to determine these two intents.

[0062] The process of determining the local intent of the first field is described below.

[0063] In some embodiments, the configured content includes multiple first fields. Determining the local intent of the first field based on the semantic information of the title and the semantic information of the first field includes: determining the associated field of each first field based on the multiple first fields; and determining the local intent of the first field by fusing the semantic information of the first field and the semantic information of the associated field of the first field.

[0064] As mentioned above, each first field has a local intent. In defining the local intent of each first field, in addition to utilizing the semantic information of the first field itself, the semantic information of its associated fields is also utilized. The semantic information of the first field itself directly reflects its local intent, while the semantic information of its associated fields indirectly supplements it. For the same first field, the expressed local intent differs depending on its associated fields. For example, for the "Frequently Used Address" field, when its associated field is "Applicant," its expressed local intent might be a home address. However, when its associated field is "Company Location," its expressed local intent might be a company address.

[0065] In some embodiments, determining the associated field of each first field based on a plurality of first fields includes: determining the association relationship between the plurality of first fields based on entities and relationships between entities in the city knowledge graph; and for each of the plurality of first fields, determining the associated field of the first field from other first fields besides the first field in the plurality of first fields based on the association relationship between the plurality of first fields.

[0066] The relationships between multiple first fields are usually implicit, meaning that the configuration personnel will not directly specify the relationships between the multiple first fields they have configured in the form. Therefore, it is necessary to analyze the relationships between multiple first fields.

[0067] After determining the relationships between multiple first fields using entities and relationships within the city knowledge graph, for each first field, the first fields that have a relationship with it are identified as its associated fields. For example, in the first heterogeneous graph, the two first fields corresponding to two nodes with an edge are each other's associated fields.

[0068] After determining the associated fields of the first field, a graph neural network model (such as a heterogeneous graph transformer) can be used to fuse the semantic information of the first field and the semantic information of its associated fields. By fusing the semantic information of the first field and its associated fields, the local intent of the first field can be determined more accurately.

[0069] Figure 3 A flowchart illustrating a partial intent for determining a first field according to some embodiments of this disclosure is shown. Figure 3 As shown, as an example of a graph neural network model, a heterogeneous graph transformer (HGT) is used to fuse the semantic information of the first field and the semantic information of the associated fields of the first field to obtain the first local intent.

[0070] The above embodiments describe the process of determining the local intent of the first field. The following describes the process of determining the overall intent of the form.

[0071] In some embodiments, the configured content includes multiple first fields. Determining the overall intent of the form based on the semantic information of the title and the semantic information of the first fields includes: for each of the multiple first fields, generating overall semantic information of the first field by combining the semantic information of the title and the semantic information of the first field; processing the overall semantic information of each first field using a graph neural network model; and determining the overall intent of the form based on the processed overall semantic information of the multiple first fields.

[0072] Since titles typically contain strong overall target semantics, concatenating the semantic information of the title with the semantic information of the first field can obtain information that integrates the overall intent of the form with the specific fields that embody that intent, thus fully exploring the overall intent of the form.

[0073] Figure 4 A flowchart illustrating the overall intent of determining a form according to some embodiments of this disclosure is shown. Figure 4 As shown, for each first field, the semantic information of the first field is concatenated with the semantic information of the title, and input into HGT (as an example of a graph neural network model). HGT is then used to perform deep semantic fusion on the concatenated information to obtain the overall semantic information of the first field. Subsequently, average pooling is performed on the overall semantic information of multiple fields to obtain the overall intent of the form.

[0074] The aforementioned processes for determining the local intent of the first field and the overall intent of the form both utilize graph neural network models. The graph neural network models used in determining both the local intent of the first field and the overall intent of the form are pre-trained. Furthermore, the aforementioned... Figures 2-4 Both [the text] used HGT as an example of a graph neural network model, which is understandable. Figures 2-4 Each HGT in the system has its own corresponding function when used, and they are not completely the same.

[0075] The overall intent of the form is determined using a first model, which is the graph neural network model used in determining the overall intent of the form. The first model is trained using first positive samples and first negative samples. The first positive samples are generated from sub-forms obtained from the same form, and the first negative samples are generated from sub-forms obtained from different forms. The first positive and first negative samples can appear as sample pairs, used to constrain the first model to make the overall intent of sub-forms from the same form as close as possible.

[0076] The local intent of the first field is determined using a second model, which is the graph neural network model used in determining the local intent of the first field. The second model is trained using second positive samples and second negative samples. The second positive samples are generated based on fields obtained from the same form, and the second negative samples are generated based on fields obtained from different forms. The second positive and second negative samples can appear as sample pairs to constrain the second model to make the local intents of fields in the same form as close as possible.

[0077] In other words, a contrastive learning strategy was used during the training of both the first and second models. By optimizing the contrastive learning objective based on the loss function setting, the models can more clearly distinguish the intent behind different forms and fields, thereby providing a high-quality and discriminative intent representation for subsequent configuration field recommendations.

[0078] Distinguishing between the overall intent of a form and the specific intent of its fields improves the adaptability of intent modeling to changes in user needs across different scenarios. After determining the overall intent of the form and the specific intent of the first field, the recommended configuration field is selected from multiple candidate fields.

[0079] This disclosure, when determining the recommended configuration fields, not only performs intent analysis on the form, but also makes full use of the form's structural information, namely the relationships between multiple fields. Furthermore, it refines the intent analysis into overall intent and local intent, which can fully explore the potential business logic of the configured content in the form, adapt to different business scenarios, and has strong recommendation accuracy.

[0080] In some embodiments, a target intent is determined based on the overall intent of the form and the local intent of the first field; a recommended value for each candidate field is determined based on the similarity between the target intent and each candidate field among multiple candidate fields; and a recommended configuration field is determined based on the recommended value of each recommended field.

[0081] The target intent combines the overall intent and the local intent, which can fully reflect the configuration logic of the current form, thereby enabling accurate recommendations for the recommended configuration fields.

[0082] In some embodiments, determining the target intent based on the overall intent of the form and the partial intent of the first field includes: determining an intermediate intent by fusing the overall intent of the form and the partial intent of the first field; generating target parameters based on the overall intent of the form; and processing the intermediate intent based on the target parameters to generate the target intent.

[0083] For example, the vector representing the overall intent of the form and the vector representing the local intent of the first field are concatenated to generate a vector representing the target intent. Then, the vector representing the target intent is concatenated with the vector of each candidate field, and the cosine similarity between the vector representing the target intent and the vector of each candidate field is calculated, serving as the recommendation value for each candidate field.

[0084] Processing intermediate intents based on target parameters can be achieved using fully connected linear layers. That is, the target parameters are the parameters of the fully connected linear layer.

[0085] After obtaining the concatenated vector representing the overall intent of the form and the concatenated vector representing the local intent of the first field, a linear transformation can be performed on the concatenated vector using a linear layer with fixed parameters to obtain a vector that can represent the target intent. The fixed parameters here are also obtained through pre-training.

[0086] Considering that the user needs for the same first field may differ in different form creation contexts, that is, the overall intent of the form reflects the form creation context, which has a strong influence on the positioning of user needs, in order to diversify form scenarios, we further consider introducing a parameter adaptive adjustment mechanism for the linear layer with fixed parameters mentioned above.

[0087] In other words, after obtaining the concatenated vector representing the overall intent of the form and the concatenated vector representing the local intent of the first field, the concatenated vector is transformed using the vector representing the overall intent of the form to obtain a vector that can represent the target intent. This allows the target intent to be dynamically adjusted according to the overall intent of different forms, thereby better adapting to diverse user needs. In other words, after determining the overall intent of the form and the local intent of the first field, the target intent can be extracted through multiple adaptively adjustable fully connected linear layers.

[0088] Figure 5 A flowchart illustrating the determination of recommended configuration fields according to some embodiments of this disclosure is shown. For example... Figure 5 As shown, the local intent of the first field and the overall intent of the form are concatenated and then input into three linear layers (these three layers are only for example). The target intent output by the last linear layer is obtained, where the target parameters of each linear layer are determined based on the overall intent of the form. Then, similarity calculations are performed between the target intent and each candidate field to determine the recommended configuration fields.

[0089] In some embodiments, generating the target parameters based on the overall intent of the form includes: generating a first matrix parameter and a second matrix parameter based on the overall intent of the form, wherein the rank of the first matrix parameter and the second matrix parameter is lower than a specified value; and generating the target parameters based on the first matrix parameter and the second matrix parameter.

[0090] The parameters of the linear layer, i.e., the target parameters, include the first matrix parameters and the second matrix parameters, and may also include pre-learned fixed weight matrix parameters. For example, for a vector A representing the overall intent of a form, two low-rank matrices B and C are generated from vector A, where the ranks of B and C are much smaller than the dimensions of their input and output vectors. The target parameter D can be expressed as D = W + B. C.

[0091] By employing a parameter adaptation mechanism driven by the overall intent of the form, the overall intent of the form can be more flexibly aligned with candidate fields, significantly improving generalization ability and accuracy in complex form construction scenarios. Furthermore, a low-rank learning method is used to generate target parameters based on the overall intent of the form, thus reducing the adjustment cost of the adaptation mechanism.

[0092] The above embodiments generate recommended configuration fields based on the content configured in the form. There can be one or more recommended configuration fields. When there are multiple recommended configuration fields, they are displayed to the user (i.e., the configuration personnel) in descending order of recommended values.

[0093] After the user confirms that the recommended configuration field is configured in the form, for example, by triggering the control to add the recommended configuration field to the form, the next recommended configuration field can be generated.

[0094] In other words, the form configuration method also includes: in response to a recommended configuration field being configured in the form, retrieving the already configured content in the form to determine the next recommended configuration field, until the form configuration is complete.

[0095] That is, the configured content in the form is retrieved again, the overall intent of the form and the field intent of the first field are redefined, and the next recommended configuration field is generated. The candidate fields used in generating the next recommended configuration field can be the candidate fields from the aforementioned embodiments, or they can be candidate fields updated based on user instructions or the configured content in the form.

[0096] Figure 6 A flowchart of a form configuration apparatus according to some embodiments of the present disclosure is shown.

[0097] like Figure 6 As shown, Figure 6 The first form configuration device 6 includes an acquisition module 61, configured to acquire the configured content in the form, wherein the configured content includes a title and a first field; and a generation module 62, configured to generate recommended configuration fields for the form, wherein the recommended configuration fields are determined from multiple candidate fields after determining the overall intent of the form and the partial intent of the first field based on the semantic information of the title and the semantic information of the first field.

[0098] In some embodiments, the configured content includes multiple first fields, and the generation module 62 is further configured to determine the associated field of each first field based on the multiple first fields; and to determine the local intent of the first field by fusing the semantic information of the first field and the semantic information of the associated field of the first field.

[0099] In some embodiments, the generation module 62 is further configured to determine the association relationship between multiple first fields based on entities and relationships between entities in the city knowledge graph; and for each of the multiple first fields, to determine the associated field of the first field from other first fields besides the first field based on the association relationship between the multiple first fields.

[0100] In some embodiments, the configured content includes multiple first fields, and the first form configuration device 6 is further configured to, for each of the multiple first fields, use at least one of a city knowledge graph and a large language model to perform semantic enhancement on the first field to obtain semantic enhancement information of the first field; determine the association relationship of the multiple first fields according to the city knowledge graph; and for each first field, process the semantic enhancement information of the first field using the association relationship of the multiple first fields to determine the semantic information of the first field.

[0101] In some embodiments, the first form configuration device 6 is further configured to perform a first semantic enhancement on the first field using a city knowledge graph to obtain first semantic enhancement information; perform a second semantic enhancement on the first field using a large language model to obtain second semantic enhancement information; and combine at least one of the first semantic enhancement information and the second semantic enhancement information with the first field to obtain semantic enhancement information of the first field.

[0102] In some embodiments, the first form configuration device 6 is further configured to obtain a first vector of the first field by fusing at least one of the first semantic enhancement information, the second semantic enhancement information, and the first field, wherein the first vector represents the semantic enhancement information of the first field; and to obtain a second vector by processing the first vector of each first field and the association relationship of multiple first fields through a graph neural network model, wherein the second vector represents the semantic information of the first field.

[0103] In some embodiments, the configured content includes multiple first fields, and the generation module 62 is further configured to generate overall semantic information of each of the multiple first fields by combining the semantic information of the title and the semantic information of the first field; process the overall semantic information of each first field using a graph neural network model; and determine the overall intent of the form based on the processed overall semantic information of the multiple first fragments.

[0104] In some embodiments, the overall intent of the form is determined using a first model trained with first positive samples and first negative samples, wherein the first positive samples are generated based on sub-forms obtained from the same form and the first negative samples are generated based on sub-forms obtained from different forms; the local intent of the first field is determined using a second model trained with second positive samples and second negative samples, wherein the second positive samples are generated based on fields obtained from the same form and the second negative samples are generated based on fields obtained from different forms.

[0105] In some embodiments, the first form configuration device 6 is further configured to determine a target intent based on the overall intent of the form and the partial intent of the first field; determine a recommended value for each candidate field based on the similarity between the target intent and each candidate field among a plurality of candidate fields; and determine a recommended configuration field based on the recommended value of each recommended field.

[0106] In some embodiments, the first form configuration device 6 is further configured to determine an intermediate intent by fusing the overall intent of the form and the partial intent of the first field; generate a target parameter based on the overall intent of the form; and process the intermediate intent based on the target parameter to generate a target intent.

[0107] In some embodiments, the first form configuration device 6 is further configured to generate a first matrix parameter and a second matrix parameter according to the overall intent of the form, wherein the rank of the first matrix parameter and the second matrix parameter is lower than a specified value; and to generate target parameters according to the first matrix parameter and the second matrix parameter.

[0108] In some embodiments, the first form configuration device 6 is further configured to, in response to a recommended configuration field being configured into the form, retrieve the already configured content in the form to generate the next recommended configuration field for the form, until the form configuration is complete.

[0109] This disclosure analyzes the overall intent of the form and the partial intent of its fields based on the content already configured in the form, and then determines the recommended configuration field from multiple candidate fields. The form configuration method of this disclosure eliminates the need for manual full form configuration; it can utilize manually configured partial content to recommend configuration fields, thus effectively improving the efficiency of form configuration.

[0110] The form configuration device in the embodiments of this disclosure can be implemented by various computing devices or computer systems, as described below. Figure 7 as well as Figure 8 Describe it.

[0111] Figure 7 A schematic diagram of the structure of a form configuration apparatus according to other embodiments of the present disclosure is shown. For example... Figure 7As shown, the second form configuration device 7 of this embodiment includes a memory 71 and a processor 72 coupled to the memory 71. The processor 72 is configured to execute the form configuration method in any of the embodiments of this disclosure based on instructions stored in the memory 71.

[0112] The memory 71 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory stores, for example, the operating system, application programs, boot loader, database, and other programs.

[0113] Figure 8 A schematic diagram of the structure of a form configuration apparatus according to some embodiments of the present disclosure is shown. For example... Figure 8 As shown, the third form configuration device 8 in this embodiment includes a memory 81 and a processor 82, which are similar to those of memory 71 and processor 72. It may also include an input / output interface 83, a network interface 84, a storage interface 85, etc. These interfaces 83, 84, 85, and the memory 81 and processor 82 can be connected, for example, via a bus 86. The input / output interface 83 provides a connection interface for input / output devices such as a display, mouse, keyboard, and touchscreen. The network interface 84 provides a connection interface for various networked devices, such as connecting to a database server or cloud storage server. The storage interface 85 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0114] Embodiments of this disclosure also provide a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements any of the aforementioned form configuration methods.

[0115] Embodiments of this disclosure also provide a computer program product including instructions that, when executed by a processor, cause the processor to perform any of the foregoing form configuration methods.

[0116] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] The above description is only a preferred embodiment of this disclosure and is not intended to limit this disclosure. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the protection scope of this disclosure.

Claims

1. A form configuration method, comprising: Retrieve the configured content from the form, wherein the configured content includes a title and a first field; The recommended configuration fields for the form are generated, wherein the recommended configuration fields are determined from multiple candidate fields after determining the overall intent of the form and the partial intent of the first field based on the semantic information of the title and the semantic information of the first field.

2. The form configuration method according to claim 1, wherein, The configured content includes multiple first fields. Determining the local intent of the first field based on the semantic information of the title and the semantic information of the first fields includes: Based on the plurality of first fields, determine the associated field for each first field; By fusing the semantic information of the first field with the semantic information of the associated fields of the first field, the local intent of the first field is determined.

3. The form configuration method according to claim 2, wherein, Based on the plurality of first fields, the associated fields for each first field are determined to include: Based on the entities in the city knowledge graph and the relationships between the entities, the association relationships among the multiple first fields are determined; For each of the plurality of first fields, based on the association relationship between the plurality of first fields, the associated field of the first field is determined from the other first fields besides the first field.

4. The form configuration method according to claim 1, wherein the configured content includes multiple first fields, and the form configuration method further includes: For each of the plurality of first fields, semantic enhancement is performed on the first field using at least one of urban knowledge graph and large language model to obtain semantic enhancement information of the first field; The relationships between the multiple first fields are determined based on the city knowledge graph; For each of the first fields, the semantic enhancement information of the first field is processed using the association relationship of the multiple first fields to determine the semantic information of the first field.

5. The form configuration method according to claim 4, wherein, Using at least one of urban knowledge graphs and large language models to semantically enhance the first field, the semantic enhancement information of the first field includes: The first field is semantically enhanced using the city knowledge graph to obtain first semantic enhancement information. The first field is then subjected to second semantic enhancement using the large language model to obtain second semantic enhancement information. At least one of the first semantic enhancement information and the second semantic enhancement information is combined with the first field to obtain the semantic enhancement information of the first field.

6. The form configuration method according to claim 5, wherein, The semantic enhancement information of the first field is processed using the association relationship of the multiple first fields to determine the semantic information of the first field, which includes: By fusing at least one of the first semantic enhancement information, the second semantic enhancement information, and the first field, a first vector of the first field is obtained, wherein the first vector represents the semantic enhancement information of the first field; A graph neural network model is used to process the first vector of each first field and the association relationship of the plurality of first fields to obtain a second vector, which represents the semantic information of the first field.

7. The form configuration method according to any one of claims 1 to 6, wherein, The configured content includes multiple first fields. Determining the overall intent of the form based on the semantic information of the title and the semantic information of the first fields includes: For each of the plurality of first fields, the overall semantic information of the first field is generated by combining the semantic information of the title and the semantic information of the first field. The overall semantic information of each first field is processed using a graph neural network model; Based on the processed overall semantic information of the multiple first fragments, the overall intent of the form is determined.

8. The form configuration method according to any one of claims 1 to 6, wherein, The overall intent of the form is determined using a first model, which is trained using a first positive sample and a first negative sample. The first positive sample is generated based on sub-forms obtained from the same form, and the first negative sample is generated based on sub-forms obtained from different forms. The local intent of the first field is determined using a second model, which is trained using a second positive sample and a second negative sample. The second positive sample is generated based on fields obtained from the same form, and the second negative sample is generated based on fields obtained from different forms.

9. The form configuration method according to any one of claims 1 to 6, further comprising: Determine the target intent based on the overall intent of the form and the local intent of the first field; Based on the similarity between the target intent and each of the plurality of candidate fields, a recommended value for each candidate field is determined; The recommended configuration field is determined based on the recommended value of each recommended field.

10. The form configuration method according to claim 9, wherein, Based on the overall intent of the form and the local intent of the first field, the target intent is determined as follows: The intermediate intent is determined by integrating the overall intent of the form and the local intent of the first field; Generate and determine target parameters based on the overall intent of the form; The intermediate intent is processed based on the target parameters to generate the target intent.

11. The form configuration method according to claim 10, wherein, Generating and determining target parameters based on the overall intent of the form includes: Generate a first matrix parameter and a second matrix parameter based on the overall intent of the form, wherein the rank of the first matrix parameter and the second matrix parameter is lower than a specified value; The target parameters are generated based on the first matrix parameters and the second matrix parameters.

12. The form configuration method according to any one of claims 1 to 6, further comprising: In response to the recommended configuration field being configured in the form, the configured content in the form is retrieved again to generate the next recommended configuration field for the form, until the form configuration is complete.

13. A form configuration device, comprising: The retrieval module is configured to retrieve the configured content from the form, wherein the configured content includes a title and a first field; The generation module is configured to generate recommended configuration fields for the form, wherein the recommended configuration fields are determined from multiple candidate fields after determining the overall intent of the form and the partial intent of the first field based on the semantic information of the title and the semantic information of the first field.

14. A form configuration device, comprising: processor; as well as A memory coupled to the processor is used to store instructions that, when executed by the processor, cause the processor to perform the form configuration method as described in any one of claims 1 to 12.

15. A computer-readable storage medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the form configuration method as described in any one of claims 1 to 12.

16. A computer program product comprising instructions that, when executed by a processor, cause the processor to perform the form configuration method as described in any one of claims 1 to 12.

Images