An industry model matching method based on natural language remote sensing information

Abstract

This invention relates to the field of natural language processing technology, and more particularly to an industry model matching method based on natural language remote sensing information. The method includes: determining the target user category based on model demand difference values ​​or the number of model demands; determining a processing strategy based on the number of high-demand users; when processing high-demand users sequentially, the processing strategy determines the processing coefficient of each label vector based on the association status of the label vectors corresponding to the high-demand users, or analyzes the matching ambiguity corresponding to the label vectors; when analyzing the task allocation method corresponding to high-demand users, the method allocates high-demand users based on the average industry matching convergence of each processing module and the stability of the supply model, using module matching or module load. This invention adaptively adjusts model matching according to actual model demand conditions in the face of complex user demands, improving model matching efficiency.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to an industry model matching method based on natural language remote sensing information. Background Technology

[0002] With the gradual maturation of natural language processing technology, in the field of remote sensing satellite mission management, user requirements are mostly submitted in the form of natural language, while industry application models need to be triggered based on clear scenarios and parameters. In the existing technology, the matching of natural language requirements and industry models relies on manual annotation or simple keyword matching, which often has the problem of single matching method and poor accuracy. Therefore, how to improve the model matching accuracy and efficiency is an urgent problem to be solved by those skilled in the art.

[0003] Chinese Patent Publication No. CN114282592B discloses a method and apparatus for industry text matching model based on deep learning. The method includes: acquiring a preset number of cross-industry data sets as a training set to obtain the sentences to be matched; inputting a deep learning-based industry text matching model NERB, and after data preprocessing, inputting it into optimized pre-trained models NEZHA, RoBERTa, and ERNIE-Gram respectively; outputting three text matching results after matching by the optimized pre-trained models; and comprehensively judging whether the output results of the industry text matching model are similar when any two or three text matching results are similar, otherwise they are dissimilar. It is evident that the above technical solution has the following problems: it cannot adaptively adjust the model matching strategy according to the actual model requirements when facing complex user needs, resulting in poor model matching efficiency. Summary of the Invention

[0004] To address this issue, the present invention provides an industry model matching method based on natural language remote sensing information, which overcomes the problem in existing technologies that fail to adaptively adjust the model matching strategy according to the actual model requirements when faced with complex user needs, resulting in poor model matching efficiency.

[0005] To achieve the above objectives, this invention provides an industry model matching method based on natural language remote sensing information, comprising: The target users are determined as low-demand users or high-demand users based on the model demand difference value or the model demand quantity. Determine the processing strategy based on the number of users with high demand; The processing strategy is to determine the coefficient analysis method based on the association state of the tag vector corresponding to the high-demand user when processing the demand sequence. The coefficient analysis method includes a first coefficient analysis method and a second coefficient analysis method. The first coefficient analysis method is to determine the processing coefficient of each label vector based on the process priority coefficient. The second coefficient analysis method is to analyze the matching ambiguity corresponding to the label vector and obtain the matching ambiguity balance value, so as to select the processing coefficient of each label vector based on the task processing time or the matching ambiguity. When analyzing the task allocation methods for high-demand users, the processing strategy is to determine the allocation of tasks to high-demand users based on module matching or module load, according to the average industry matching closeness of each processing module and the stability of the supply model. Periodically determine whether to perform cache adjustment processing based on the frequency of retrieval corresponding to the industry model.

[0006] Furthermore, for target users whose model demand difference value is less than or equal to the preset model demand difference value or whose model demand quantity is less than or equal to the preset model demand quantity, the target user is recorded as a low demand user. Furthermore, based on the condition that the number of high-demand users is greater than the preset number of high-demand users, the demand order is processed for the tag vectors corresponding to high-demand users; Furthermore, the demand sequence processing includes detecting the association status of the tag vectors corresponding to high-demand users; If the number of users in the first associated state is greater than the preset number of users in the first associated state, then the processing coefficient of each workflow association vector is determined according to the process priority coefficient. If the number of users in the first associated state is less than or equal to the preset number of users in the first associated state, then the matching ambiguity corresponding to each non-associated vector is analyzed.

[0007] The first associated user is a high-demand user whose workflow association vector percentage in the corresponding tag vector is greater than the preset workflow association vector percentage.

[0008] Furthermore, when analyzing the matching ambiguity corresponding to the tag vector, the matching ambiguity balance value corresponding to the non-associated vector of high-demand users is detected. If the matching ambiguity balance value is greater than the preset matching ambiguity balance value, the processing coefficient of each non-associated vector is determined according to the task processing time. If the matching fuzziness balance value is less than or equal to the preset matching fuzziness balance value, then the processing coefficients of each non-associated vector are determined based on the matching fuzziness. Furthermore, the processing coefficients corresponding to non-associated vectors are positively correlated with the matching ambiguity and the task processing time.

[0009] Furthermore, the analysis of task allocation methods for high-demand users includes: The average industry matching convergence and supply model stability of each processing module are detected. If the average industry matching convergence is less than the preset average industry matching convergence or the supply model stability is less than the preset supply model stability, then allocation is made to high-demand users based on module load. If the industry matching mean is greater than or equal to the preset industry matching mean and the supply model stability is greater than or equal to the preset supply model stability, then allocation will be made to high-demand users based on module matching.

[0010] Furthermore, for industry models where the frequency of accessing trending topics exceeds the preset frequency, cache adjustment processing is determined. Furthermore, the cache adjustment process includes increasing the cache duration of the industry model based on the frequency of retrieval. The increase in cache duration is positively correlated with the frequency of retrieval.

[0011] Compared with the prior art, the beneficial effect of the present invention is that the technical solution of the present invention reflects the diversity of target user needs through model demand difference value and reflects the number of user needs through model demand quantity, and classifies target users accordingly, thereby achieving effective user segmentation and improving the accuracy of subsequent determination of processing strategy based on the number of high demand users.

[0012] Furthermore, in the technical solution of the present invention, the number of users in the first associated state reflects the proportion of users with a large proportion of workflow association vectors, and the corresponding selection determines the processing coefficient of each workflow association vector based on the workflow priority coefficient or analyzes the matching ambiguity corresponding to each non-associated vector, making the determination of the processing coefficient more targeted, thereby improving data processing efficiency and thus improving the matching efficiency of industry models.

[0013] Furthermore, in the technical solution of this invention, the matching fuzziness balance value of the non-associated vectors of high-demand users reflects the difficulty of matching the non-associated vectors of high-demand users with the industry model. And according to the actual situation of the matching fuzziness balance value, the non-associated vector processing coefficient is determined according to the task processing time or the matching fuzziness, which improves the setting accuracy of the processing coefficient and thus improves the processing speed of high-demand users.

[0014] Furthermore, in the technical solution of this invention, the industry matching proximity mean reflects the similarity of the industry fields of the industry models cached by the processing module, and the supply model stability reflects the degree of change of the industry fields of the industry models cached by the processing module. Different allocation methods for high-demand users are selected accordingly, so that the allocation of high-demand users is more in line with the actual application scenario, thereby improving the industry model matching speed. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the industry model matching method based on natural language remote sensing information according to the present invention; Figure 2 This is a flowchart illustrating how the present invention determines the processing strategy based on the number of high-demand users; Figure 3 This is a flowchart illustrating the correlation state determination coefficient analysis method based on the tag vectors corresponding to high-demand users in this invention. Detailed Implementation

[0016] To make the objectives and advantages of this invention clearer, the invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0017] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0018] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0019] Please see Figures 1 to 3 As shown, this invention provides an industry model matching method based on natural language remote sensing information, including: The target users are determined as low-demand users or high-demand users based on the model demand difference value or the model demand quantity. Determine the processing strategy based on the number of users with high demand; The processing strategy is to determine the coefficient analysis method based on the association state of the tag vector corresponding to the high-demand user when processing the demand sequence. The coefficient analysis method includes a first coefficient analysis method and a second coefficient analysis method. The first coefficient analysis method is to determine the processing coefficient of each label vector based on the process priority coefficient. The second coefficient analysis method is to analyze the matching ambiguity corresponding to the label vector and obtain the matching ambiguity balance value, so as to select the processing coefficient of each label vector based on the task processing time or the matching ambiguity. When analyzing the task allocation methods for high-demand users, the processing strategy is to determine the allocation of tasks to high-demand users based on module matching or module load, according to the average industry matching closeness of each processing module and the stability of the supply model. Periodically determine whether to perform cache adjustment processing based on the frequency of retrieval corresponding to the industry model.

[0020] This invention is specifically applied to industry model matching, specifically to a remote sensing satellite mission platform. This platform includes several processing modules with data processing capabilities and a cloud platform storing industry models. Specifically, users can upload requirement text. The processing modules use a BERT model to process the requirement text, obtaining several label vectors. Model matching is then performed on each label vector, and the successfully matched industry model is retrieved from the cloud platform and loaded into the processing module for user use. For example, if the user uploads the requirement text "monitoring wheat growth in a certain area," the corresponding feedback industry model is a pre-trained model for observing crop growth. The text processing for the requirement text includes sentence segmentation, word segmentation, semantic cleaning, and entity recognition (entity recognition identifies key entities, including but not limited to "region"). The system identifies key entities ("crop type" and "monitoring task") to generate several sub-requirements. These sub-requirements are then converted into vector representations using Sentence-BERT, and these vectors are denoted as tag vectors. Since a single requirement text may contain multiple sub-requirements, the number of corresponding tag vectors is not fixed at one. For example, if the requirement text is "I want to simultaneously monitor the growth and pest and disease situation of winter wheat in the North China Plain, and the drought situation of corn in Northeast China," the corresponding sub-requirements include: 1. Monitoring the growth of winter wheat (crop type) in the North China Plain (region) (monitoring task); 2. Monitoring the pest and disease situation of winter wheat (crop type) in the North China Plain (region) (monitoring task); 3. Monitoring the drought situation of corn (crop type) in Northeast China (region) (monitoring task). The above content is easily understood by those skilled in the art and will not be elaborated upon here. The target user is the user who uploaded the requirement text within the most recent processing cycle. The processing cycle is a continuous cycle with a duration of 1 hour. The greater the user's data acquisition requirement within the processing cycle, the larger the value of the processing cycle.

[0021] The matching process parameter statistics are records of detection parameters that meet the processing requirements during the historical use of the remote sensing satellite mission platform. The detection parameters include model requirement difference value, number of model requirements, number of users with high requirements, proportion of workflow association vector, number of users in the first association state, ambiguity balance value, task processing time, matching ambiguity, industry matching closeness, supply model stability, and retrieval frequency. Meeting the processing requirements means that the model matching accuracy during the corresponding use process meets the user's needs. This is content that is already known to those skilled in the art and will not be elaborated here.

[0022] Specifically, for target users whose model demand difference value is less than or equal to the preset model demand difference value or whose model demand quantity is less than or equal to the preset model demand quantity, the target user is recorded as a low demand user. For target users whose model demand difference value is greater than the preset model demand difference value and whose model demand quantity is greater than the preset model demand quantity, the target user is recorded as a high demand user.

[0023] For a single target user, the method for confirming the corresponding model requirement difference value is to detect the proportion of non-identical items corresponding to each key entity. The model requirement difference value = 1 - the average of the proportions of identical items corresponding to each key entity. For a single key entity, the proportion of identical items is the number of identical items corresponding to that key entity in the most recent requirement text for that target user / the total number of key entities. For example, for the key entity "region", if "North China Plain" appears twice and "Northeast Plain" once in the sub-requirements in the most recent requirement text, then the number of identical items is 2, the total number of key entities is 3, and the total number of key entities is the total number of occurrences, so the proportion of identical items = 2 / 3. The number of model requirements reflects the total number of industry models provided to the target user after retrieval and matching to solve the target user's needs. The number of model requirements is the total number of matched industry models.

[0024] Regarding the preset model requirement difference value and preset model requirement quantity, the model requirement difference value reflects the diversity of models required by the target user's corresponding requirement text, and the model requirement quantity reflects the total number of industry models required. This allows for the division of users into high-demand users with multiple model requirements and low-demand users with no multiple model requirements. Therefore, the higher the user's judgment benchmark for multiple model requirements corresponding to high-demand users, the larger the values ​​of the preset model requirement difference value and preset model requirement quantity. A value determination method is provided, which extracts the model requirement difference value and model requirement quantity from the parameter statistics of the matching process, and records the average values ​​of the model requirement difference value and model requirement quantity as the preset model requirement difference value and preset model requirement quantity, respectively. Outlier removal can be performed on the model requirement difference value and model requirement quantity. How to perform outlier removal is a matter known to those skilled in the art and does not need to be elaborated.

[0025] Specifically, based on the condition that the number of high-demand users is greater than the preset number of high-demand users, the demand order is processed for the tag vectors corresponding to high-demand users.

[0026] The number of high-demand users is the total number of high-demand users in the most recent processing cycle. The preset value for the number of high-demand users is understood to be that a larger number of high-demand users in the most recent processing cycle reflects a greater number of users with multi-model needs, thus indicating value for demand-sequential processing. Therefore, the greater the user's demand for the contribution of high-demand users in the most recent processing cycle to the analytical value of demand-sequential processing, the larger the preset value for the number of high-demand users. One method for setting this value is to extract the matching process parameters to count the corresponding number of high-demand users, and then record the average value of the number of high-demand users as the preset number of high-demand users.

[0027] One specific implementation provides a set of values, wherein the total number of target users in the most recent processing cycle is 100, and the preset total number of high-demand users is 75.

[0028] Specifically, demand order processing includes detecting the association status of the tag vectors corresponding to users with high demand; If the number of users in the first associated state is greater than the preset number of users in the first associated state, then the processing coefficient of each workflow association vector is determined according to the process priority coefficient. If the number of users in the first associated state is less than or equal to the preset number of users in the first associated state, then the matching ambiguity corresponding to each non-associated vector is analyzed.

[0029] The first associated user is a high-demand user whose workflow association vector percentage in the corresponding tag vector is greater than the preset workflow association vector percentage.

[0030] In the tag vectors corresponding to a single high-demand user, for a single tag vector, if the tag vector is associated with at least one other tag vector through a workflow, then the tag vector is denoted as a workflow-associated vector; if the tag vector is not associated with any other tag vector through a workflow, then the tag vector is denoted as a non-associated vector. The proportion of workflow-associated vectors = the number of workflow-associated vectors corresponding to the high-demand user / the total number of tag vectors corresponding to the high-demand user. For any two tag vectors, the industry models corresponding to each of the two tag vectors are obtained respectively. If the output of one industry model is the same as that of the other industry model... If the input is a label vector, then the two label vectors are associated in a workflow. If the output of an industry model A is the input of another industry model B, then the label vector corresponding to industry model A is the priority input vector. For example, one model is a "pest and disease monitoring model" whose output is the risk assessment level of pests and diseases in a certain area. Another model is a "pest and disease control strategy recommendation model" which recommends pest and disease control strategies based on the risk assessment level of pests and diseases output by the "pest and disease monitoring model". Then the label vectors corresponding to the "pest and disease monitoring model" and the "pest and disease control strategy recommendation model" are associated in a workflow, and the label vector corresponding to the "pest and disease monitoring model" is the priority input vector.

[0031] The values ​​of the preset workflow association vector ratio and the preset number of users in the first association state are understood to reflect the following: a larger workflow association vector ratio indicates a larger number of workflow association vectors in the tag vectors corresponding to high-demand users; a larger number of users in the first association state indicates a stronger reliability in the contribution of the current number of users in the first association state to the processing coefficients determined based on the workflow priority coefficient. Therefore, when the number of users in the first association state is greater than the preset number of users in the first association state, the processing coefficient is determined based on the workflow priority coefficient for the workflow association vectors. This avoids the problem of poor data reliability caused by choosing to determine the processing coefficient based on the workflow priority coefficient when the number of workflow association vectors in the tag vector is too small. Therefore, the higher the user's requirement for the accuracy of the processing coefficients determined based on the workflow priority coefficient, the larger the values ​​of the preset workflow association vector ratio and the preset number of users in the first association state. One method for setting these values ​​is to extract the matching process parameters and statistically analyze the corresponding workflow association vector ratio and the number of users in the first association state. Outliers in the workflow association vector ratio and the number of users in the first association state are then removed. The average values ​​of the workflow association vector ratio and the number of users in the first association state after removing outliers are recorded as the preset workflow association vector ratio and the preset number of users in the first association state, respectively.

[0032] When determining the processing coefficient of a workflow association vector based on the workflow priority coefficient, the confirmation method for the corresponding workflow priority coefficient for a single workflow association vector is as follows: extract a number of historical samples corresponding to the workflow association vector, detect the number of times the workflow association vector is used as a priority input vector in the historical samples, and record the number of times it is used as a priority input vector as the workflow priority coefficient corresponding to the workflow association vector. The larger the value of the number of historical samples, the higher the accuracy of the corresponding workflow priority coefficient determination. One possible value is 20 historical samples.

[0033] For a workflow association vector, its corresponding historical sample is the requirement text that is also a workflow association vector in the most recent cycle of processing.

[0034] Specifically, when analyzing the matching ambiguity corresponding to non-associated vectors, the ambiguity balance value of the matching ambiguity corresponding to the non-associated vectors of high-demand users is detected. If the ambiguity balance value is greater than the preset ambiguity balance value, the processing coefficient of each non-associated vector is determined according to the task processing time. If the matching fuzziness balance value is less than or equal to the preset matching fuzziness balance value, then the processing coefficients of each non-associated vector are determined based on the matching fuzziness.

[0035] For a single unrelated vector, the corresponding matching ambiguity is determined by detecting the maximum matching degree of that unrelated vector. The matching ambiguity is calculated as 1 - maximum matching degree. The formula for calculating the matching ambiguity balance value S is as follows: ; in, For the first Matching ambiguity corresponding to each non-associated vector This represents the average matching ambiguity for each non-associated vector. This represents the total number of non-associated vectors. To match the fuzziness balance value, The value of is an integer. Each industry model in this invention corresponds to a model label. For example, the livestock quantity statistics model has model labels including remote sensing monitoring, computer vision, multimodal fusion, livestock quantity monitoring, and high-resolution processing. For a single industry model, a model label vector is generated based on its corresponding model label. How to fuse labels to generate vector representations is a topic already understood by those skilled in the art and will not be elaborated further. When performing model matching on any unrelated vector or workflow-related vector, the industry model with the highest cosine similarity to the corresponding model label vector is recorded as a successful match, and this cosine similarity is recorded as the maximum matching degree.

[0036] Regarding the preset matching fuzziness balance value, this value reflects the user's acceptance of the model matching results. The smaller the preset matching fuzziness balance value, the more closely the matching result of the unrelated vector matches the target user's needs. A method for determining this value is provided: extracting the average of the matching fuzziness balance values ​​corresponding to the statistical analysis of the matching process parameters and recording it as the preset matching fuzziness balance value.

[0037] Specifically, if the number of users in the first associated state is greater than the preset number of users in the first associated state, the processing module processes the workflow associated vectors in descending order of processing coefficients, and processes the non-associated vectors in a random order. If the number of users in the first associated state is less than or equal to the preset number of users in the first associated state, non-associated vectors are processed in descending order of processing coefficients, while workflow associated vectors are processed in a random order.

[0038] Specifically, the processing coefficients of non-associated vectors are positively correlated with the matching ambiguity and the task processing time.

[0039] When determining the processing coefficient of an unrelated vector based on task processing time, the processing coefficient corresponding to the unrelated vector = preset task processing time / task processing time. When determining the processing coefficient of an unrelated vector based on task processing time, the processing coefficient corresponding to the unrelated vector = matching fuzziness / preset matching fuzziness. For a single unrelated vector, its corresponding task processing time is the average of the reference usage times of the industry model corresponding to that unrelated vector. For a single industry model, the user can input the model themselves and detect the time of the input result of that industry model, which is recorded as a reference usage time. The larger the number of extracted reference usage times, the greater the accuracy of the task processing time. One possible value is provided: the number of reference usage times is 10. It's easy to see that the longer the task processing time and the greater the matching ambiguity, the more difficult it is to process the corresponding unrelated vectors, and the more time is required. Therefore, the higher the user's demand for the processing efficiency of unrelated vectors, the smaller the values ​​of the preset task processing time and the preset matching ambiguity. One possible value is to extract the task processing time and matching ambiguity from the statistical parameters of the matching process, and record the average values ​​of the task processing time and the matching ambiguity as the preset task processing time and the preset matching ambiguity.

[0040] Specifically, based on the condition that the number of high-demand users is less than or equal to the number of users in a preset category, the task allocation method for high-demand users is determined and analyzed.

[0041] Specifically, the analysis of task allocation methods for high-demand users includes: The average industry matching convergence and supply model stability of each processing module are detected. If the average industry matching convergence is less than the preset average industry matching convergence or the supply model stability is less than the preset supply model stability, then allocation is made to high-demand users based on module load. If the industry matching mean is greater than or equal to the preset industry matching mean and the supply model stability is greater than or equal to the preset supply model stability, then allocation will be made to high-demand users based on module matching.

[0042] For a single processing module, the method for confirming its corresponding industry matching closeness is as follows: detect the processing demand text processed by the module in the most recent cycle and the corresponding matching industry model. The maximum number of industry models in the same industry field is recorded as the industry matching closeness. The average industry matching closeness is the average value of the industry matching closeness. The industry field applied to the industry model is manually labeled by the user. For example, the industry field includes, but is not limited to, agriculture, fishery and animal husbandry. This is content that is already known to those skilled in the art and will not be elaborated here. For the stability of the supply model, obtain the absolute value of the difference between the maximum and minimum values ​​of the industry matching closeness corresponding to the most recent 5 cycles of processing for each processing module. The average value of the absolute values ​​corresponding to each processing module is recorded as the stability of the supply model.

[0043] The module matching-based allocation to high-demand users includes: detecting the industry models with the largest number of identical industry fields corresponding to each processing module, which are denoted as the popularity fields of that processing module; filtering processing modules that are identical to the popularity fields corresponding to high-demand users; and selecting the processing module with the largest number of values ​​to process the demand text of high-demand users. The popularity fields corresponding to high-demand users are the industry fields that appear most frequently in the industry models matched in the history of high-demand users.

[0044] The system provides a preset industry matching proximity and a preset supply model stability, offering a set of values. These values ​​are extracted from the statistical parameters of the matching process. The average values ​​of these two values ​​are then recorded as the preset industry matching proximity and preset supply model stability. It is evident that the higher the values ​​of these two values, the more stable and similar the domain of the processing module's target text becomes. Therefore, the preset industry matching proximity and preset supply model stability can be set according to actual needs.

[0045] When allocating resources to high-demand users based on module load, the allocation is done in a random order. When allocating resources to a single high-load user, the user is assigned to the processing module with the lowest module load. Module load is the current CPU utilization of the processing module.

[0046] Specifically, for industry models where the frequency of accessing trending topics exceeds the preset frequency, cache adjustment processing will be implemented.

[0047] Specifically, the cache adjustment process includes increasing the cache duration of the industry model based on the frequency of retrieval. The increase in cache duration is positively correlated with the frequency of retrieval.

[0048] Specifically, the processing module has a caching mechanism. If the processing module processes a demand text and successfully matches it to obtain the corresponding industry model, it will cache the model in the processing module for a base cache duration. After the base cache duration, the corresponding industry model will be deleted from the processing module. In the cache adjustment process, the adjusted cache duration = base cache duration × (retrieval frequency / preset retrieval frequency). For a single industry model, the corresponding retrieval frequency is the number of times the industry model was retrieved in the most recent processing cycle.

[0049] Specifically, the baseline cache duration is 30 minutes. It can be understood that the less tolerant users are of the load redundancy of the processing module, the shorter the baseline cache duration will be. The preset call frequency is 10. A value selection method is provided, which extracts the call frequency from the statistics of the matching process parameters and records the average value of the call frequency as the preset call frequency.

[0050] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An industry model matching method based on natural language remote sensing information, characterized in that, include: The target users are determined as low-demand users or high-demand users based on the model demand difference value or the model demand quantity. Determine the processing strategy based on the number of users with high demand; The processing strategy is to determine the coefficient analysis method based on the association state of the tag vector corresponding to the high-demand user when processing the demand sequence. The coefficient analysis method includes a first coefficient analysis method and a second coefficient analysis method. The first coefficient analysis method is to determine the processing coefficient of each label vector based on the process priority coefficient. The second coefficient analysis method is to analyze the matching ambiguity corresponding to the label vector and obtain the matching ambiguity balance value, so as to select the processing coefficient of each label vector based on the task processing time or the matching ambiguity. When analyzing the task allocation methods for high-demand users, the processing strategy is to determine the allocation of tasks to high-demand users based on module matching or module load, according to the average industry matching closeness of each processing module and the stability of the supply model. Periodically determine whether to perform cache adjustment processing based on the frequency of retrieval corresponding to the industry model.

2. The industry model matching method based on natural language remote sensing information according to claim 1, characterized in that, For target users whose model demand difference value is less than or equal to the preset model demand difference value or whose model demand quantity is less than or equal to the preset model demand quantity, the target user is recorded as a low demand user. For target users whose model demand difference value is greater than the preset model demand difference value and whose model demand quantity is greater than the preset model demand quantity, the target user is recorded as a high demand user.

3. The industry model matching method based on natural language remote sensing information according to claim 2, characterized in that, Based on the condition that the number of high-demand users is greater than the preset number of high-demand users, the demand order is processed for the tag vectors corresponding to high-demand users.

4. The industry model matching method based on natural language remote sensing information according to claim 3, characterized in that, Demand sequence processing includes detecting the association status of the tag vectors corresponding to users with high demand; If the number of users in the first associated state is greater than the preset number of users in the first associated state, then the processing coefficient of each workflow association vector is determined according to the process priority coefficient. If the number of users in the first associated state is less than or equal to the preset number of users in the first associated state, then the matching ambiguity corresponding to each non-associated vector is analyzed. The first associated user is a high-demand user whose workflow association vector percentage in the corresponding tag vector is greater than the preset workflow association vector percentage.

5. The industry model matching method based on natural language remote sensing information according to claim 4, characterized in that, When analyzing the matching ambiguity corresponding to the tag vector, the matching ambiguity balance value corresponding to the tag vector of high-demand users is detected. If the matching ambiguity balance value is greater than the preset matching ambiguity balance value, the processing coefficient of each tag vector is determined according to the task processing time. If the matching ambiguity balance value is less than or equal to the preset matching ambiguity balance value, then the processing coefficients of each label vector are determined based on the matching ambiguity.

6. The industry model matching method based on natural language remote sensing information according to claim 4 or 5, characterized in that, The processing coefficients for non-associated vectors are positively correlated with the matching ambiguity and the task processing time.

7. The industry model matching method based on natural language remote sensing information according to claim 2, characterized in that, Based on the condition that the number of high-demand users is less than or equal to the number of users in the preset category, the task allocation method for high-demand users is determined and analyzed.

8. The industry model matching method based on natural language remote sensing information according to claim 7, characterized in that, The analysis of task allocation methods for high-demand users includes: The average industry matching convergence and supply model stability of each processing module are detected. If the average industry matching convergence is less than the preset average industry matching convergence or the supply model stability is less than the preset supply model stability, then allocation is made to high-demand users based on module load. If the industry matching mean is greater than or equal to the preset industry matching mean and the supply model stability is greater than or equal to the preset supply model stability, then allocation will be made to high-demand users based on module matching.

9. The industry model matching method based on natural language remote sensing information according to claim 1, characterized in that, For industry models where the frequency of accessing trending topics exceeds the preset frequency, cache adjustment processing will be implemented.

10. The industry model matching method based on natural language remote sensing information according to claim 9, characterized in that, The cache adjustment process includes increasing the cache duration of the industry model based on the frequency of retrieval. The increase in cache duration is positively correlated with the frequency of retrieval.

Images