A power grid intelligent question-answering system based on question information missing completion

Abstract

The present invention relates to a power grid intelligent question-answering system based on question information missing completion, comprising a professionalism calculation module, a missingness analysis module and a semantic support module, and respectively configuring a first response module, a second response module and a third response module at the application layer, a knowledge layer and a data layer, classifying professionalism by user information and the text of the user's question, and calling different modules according to the professionalism classification to complete the completion of the question information, so that all questions can obtain a clearer semantic interpretation while performing targeted identification of the target question, thereby improving the accuracy of question parsing, making the intelligent question-answering system applicable to the field of power grid information question processing, and making professional terms more easily recognized by the semantic recognition algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid question answering, and more particularly to a power grid intelligent question answering system based on missing question information completion. Background Art

[0002] At present, with the popularization of smart grids, more and more grid data are being collected, stored, analyzed, and output. With data analysis as the basis, more and more grid services can be provided through intelligent data systems, taking into account the electricity consumption, distribution, and energy conservation requirements of various regions, scenarios, and types. The functions of smart grids are not limited to this. More smart grids have realized the provision of intelligent data services to corresponding users, making user data traceable, data retrieval permissions configurable, data dimensions more comprehensive, and data correlation higher. However, it is found during use that more user needs are still unformatted, so human support will still be required. The current design method is to understand the user's unformatted language and retrieve the corresponding grid data through offline and online manual services. , which results in still high manual service costs, and errors depend on personnel management. The authorization announcement number CN110909140B discloses a central routing method and device that combines graph question answering with retrieval question answering. By building a knowledge graph and combining semantic recognition of users, user questions are answered, and customer intentions and semantic recognition are judged by parsing the questions. However, this method is only applicable to the needs of ordinary users. At present, the demand for power grid information acquisition is the demand for information retrieval related to power grid development planning. Due to the problem of information granularity, this intelligent question answering technology still has great difficulties in semantic recognition in professional fields. Summary of the Invention

[0003] In view of this, the object of the present invention is to provide a power grid intelligent question-answering system based on missing question information completion.

[0004] In order to solve the above technical problems, the technical solution of the present invention is: a power grid intelligent question-answering system based on question information missing completion, including application layer, support layer, knowledge layer and data layer,

[0005] The support layer is configured with a professionalism calculation module, a missing analysis module, and a semantic support module. The professionalism calculation module calculates the expected professionalism value of the target question based on user information and text information;

[0006] The missing analysis module includes a format matching unit and a missing identification unit. The format matching unit is configured with a standard format database, which is configured with a number of standard format data. The format matching unit matches the corresponding standard format data according to the user information and the text information. The missing identification unit identifies missing feature data in the text information based on the standard format data.

[0007] The semantic support module is configured with a first support unit, a second support unit, and a third support unit. The semantic support module calls the first support unit, the second support unit, or the third support unit according to whether the professional expectation value falls within a preset benchmark expectation range; the first support unit outputs a first missing instruction to the application layer according to the missing feature data, the second support unit outputs a second missing instruction to the knowledge layer according to the missing feature data, and the third support unit outputs a third missing instruction to the data layer according to the missing feature data;

[0008] The application layer is configured with a first response module, which generates missing selection information according to the first missing instruction and outputs it to wait for the user to input supplementary information;

[0009] The knowledge layer is configured with a second response module, the second response module is configured with a knowledge format library, the knowledge format library stores a plurality of graph index formats, the second response module matches the graph index format according to the second missing instruction, and generates supplementary information according to the graph index format;

[0010] The data layer is configured with a third response module, and the third response module is configured with an industry terminology database. The industry terminology database is configured with a number of industry terms, and each industry term is configured with associated index information. The third response module retrieves the corresponding industry term according to the matching of the third missing instruction and the associated index information to generate supplementary information.

[0011] Furthermore, the professionalism calculation module includes an information acquisition unit and a text acquisition unit;

[0012] The information acquisition unit is used to acquire user information, the user information including access port data and user number data, the access port data reflects the address of the user access terminal, and the user number data reflects the account used by the user;

[0013] The text acquisition unit is used to convert the target question into text information;

[0014] The professional calculation module is associated with a port database, an interaction database, and a professional feature vocabulary, and is configured with an expected calculation strategy. The expected calculation strategy includes

[0015] Step A1: acquiring corresponding port configuration data from a port database according to the access port data and acquiring corresponding user interaction data from an interaction database according to the user number data;

[0016] Step A2: identifying a type of professional term in a professional feature vocabulary corresponding to the text information;

[0017] Step A3: Calculate the expected value of professionalism.

[0018] Where A is the expected value of professionalism, α c1 is the port reliability coefficient in the port configuration data, a c1 is the topology reliability coefficient corresponding to the port configuration data, Δa is the preset topology benchmark parameter, α c2 is the corresponding user reliability coefficient in the user interaction data, a c2 is the corresponding effective response rate in the user interaction data, β is the preset frequency weight parameter, and f t is the time-weighted interaction frequency in the user interaction data, Δf is the average interaction frequency in the user interaction data, and a c3 is the correct response rate in user interaction data, α c3 is the preset term reliability coefficient, e n is the professional weight value corresponding to the nth first-class professional term in the target question, where n is the number of first-class professional terms in the target question.

[0019] Furthermore, the missing analysis module includes a part-of-speech extraction unit, which extracts corresponding part-of-speech items according to the knowledge element graph and generates the standard format data according to the part-of-speech items. The part-of-speech items reflect the requirements for words in the standard format data.

[0020] Furthermore, the format matching unit is configured with a format matching strategy, and the format matching strategy includes

[0021] Step B1: configure a first matching threshold, calculate the matching result between the text information and each standard format data to obtain a first matching value, select the standard format data whose first matching value is greater than the first matching threshold to generate a first matching data group, and χ1=(∑g k ) / G, where k=1, 2, ... i, χ1 is the first matching value, g k is the part-of-speech weight of the kth part-of-speech item that matches successfully, i is the number of part-of-speech items that match successfully, and G is the sum of the part-of-speech weights of all part-of-speech items in the standard format data;

[0022] Step B2: Match the unmatched part-of-speech items of the standard format data in the first matching group with the user information to obtain a second matching value, and χ2=(∑h k ) / G, where k=1, 2, ... j, χ2 is the second matching value, h k is the part-of-speech weight of the kth part-of-speech item that matches successfully, and j is the number of part-of-speech items that match successfully;

[0023] Step B3: Calculate the weighted matching value of each standard format data in the first matching data group, with x s=d s χ1+χ2, where χ s is the weighted matching value, d s The preset matching weight parameter is used to determine the standard format data with the highest weighted matching value as the matching result.

[0024] Furthermore, the missing identification unit obtains the part-of-speech items that are not successfully matched in the standard format data in step B1, and generates missing feature data according to the part-of-speech items and text information.

[0025] Furthermore, the first supporting unit includes extracting a missing feature index from text information, the first missing instruction is generated based on the missing feature index, the first response module is associated with a selection index database, the selection index database pre-stores a number of missing selection information, and the corresponding missing selection information is retrieved based on the missing feature index.

[0026] Furthermore, the second response module is also configured with a standard text database, which stores a number of standard text data, retrieves the corresponding standard text data according to the obtained atlas index format, and obtains the difference data between the standard text data and the text information to generate the supplementary information.

[0027] Furthermore, the second response module further includes a first coordination unit, and the first coordination unit is configured to generate missing selection information according to the difference data and output it to the application layer when a plurality of difference data are generated simultaneously.

[0028] Furthermore, the third supporting unit includes extracting a missing feature index from the text information, generating the third missing instruction based on the missing feature index, and calculating the supplementary priority value of each industry term related to the missing feature index based on the associated index information. Where k = 1, 2, ... x, Q is the supplementary priority value, δ1 is the preset frequency priority parameter, q1 is the first frequency in the associated index information, which reflects the total number of times the industry term appears in different databases of the data layer, u x is the database priority corresponding to the x-th database in the data layer, t0 is the current time, t x is the last update time corresponding to the x-th database, q2 is the second frequency in the associated index information, δ2 is a preset associated priority parameter, the second frequency reflects the number of times the industry term is used in the application layer, and δ3 is the baseline priority parameter in the associated index information; the third response module selects the supplementary priority value that meets the first preset condition to generate the supplementary information.

[0029] Furthermore, the first preset condition is configured with a first recognition threshold and a second recognition threshold. When the supplementary priority value corresponding to the industry term is greater than the first supplementary threshold, or the supplementary priority value corresponding to the industry term is greater than the second recognition threshold of the average priority value, the first recognition condition is considered to be met. The average priority value is the average of the supplementary priority values ​​of all industry terms.

[0030] The third response module includes a second coordination unit, which is configured to generate missing selection information based on the industry terms and output it to the application layer when there are multiple industry terms that meet the first preset condition at the same time.

[0031] The technical effects of the present invention are mainly reflected in the following aspects: professionalism is classified through user information and the text of the user's question, and different modules are called according to the professionalism classification to complete the question information, so that all questions can be given a clearer semantic interpretation while the target questions are identified in a targeted manner. In this way, the accuracy of question parsing is improved, so that intelligent question answering can be applied to the field of power grid information question processing, and professional terms can be more easily recognized by semantic recognition algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 : The schematic diagram of the architecture of a power grid intelligent question-answering system based on question information missing completion in the present invention;

[0033] Figure 2 : The present invention provides a module topology diagram of a power grid intelligent question-answering system based on missing question information completion. DETAILED DESCRIPTION

[0034] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings to make the technical solutions of the present invention easier to understand and grasp.

[0035] Reference Figure 1As shown, a power grid intelligent question-answering system based on missing information completion for question sentences includes an application layer, a support layer, a knowledge layer, and a data layer. Generally speaking, existing question-answering systems also have a four-layer architecture. The application layer's main function is to enable user interaction, such as questions and answers about policies, regulations, standards, system functions, and business indicators, as well as in-depth analysis of energy policies and development standards, and then to run monitoring and document framework definition and maintenance. The support layer provides services for the application layer, such as semantic parsing services to parse user questions, information extraction services to extract information based on parsing results, image content extraction services to extract information based on input images, text analysis services to analyze content based on uploaded files, and knowledge reasoning services to infer corresponding elements based on the obtained content. The support layer also provides basic services such as messaging, workflow, security, permissions, and monitoring. The data layer mainly provides the data foundation, such as news, consultation, system standards, analysis reports, topological data, project data, map data, policies and regulations, experience cases, professional libraries, training videos, electricity data, archival data, practical data, etc.; the knowledge layer mainly extracts data, establishes a basic knowledge base, and performs entity recognition, entity attribute classification, entity relationship combing, attribute relationship combing; as well as concept definition, attribute definition, relationship definition, constraint rule setting and entity classification; and provides support for intelligent question and answer through algorithm models. The core of the present invention is that when the user asks a question at the application layer, the question is parsed, and then the user level is identified through the support layer, the question is completed, and semantic understanding is performed to provide technical support for the user's target question to find the answer from the knowledge graph.

[0036] The support layer is configured with a professional calculation module, a missing analysis module, and a semantic support module. The core logic of the present invention is that when a user inputs a question, a corresponding professional estimate is generated, and then the missing data is analyzed and supplemented by the semantic support module to provide more reliable support for semantic recognition.

[0037] The professionalism calculation module calculates the expected professionalism value of the target question based on user information and text information; the professionalism calculation module includes an information acquisition unit and a text acquisition unit; first, the user enters information into a text box or uploads a picture to ask a question, and the present invention includes two parts, one is the visible part of the text, which is obtained through the text acquisition unit, and the other part is implicit information, such as the user's address or power grid department. For example, if the user belongs to A1, then when asking the question, if the place name is not mentioned, it may be that the place name is ignored by the habit of use, and these terms can be supplemented by hidden information.

[0038] The information acquisition unit is used to obtain user information, and the user information includes access port data and user number data. The access port data reflects the address of the user's access terminal, and the user number data reflects the account used by the user; the access port data includes the IP address and the access method. The location is analyzed by the IP address, including whether the access is through a mobile terminal or a fixed terminal, including whether the port is an internal port and the number to which it belongs. Other information of the user can be analyzed. Preferably, a configuration database is established for the internal port, so that the port can be marked to determine the user's affiliation and the user's professional situation by the address. On the other hand, the user's information such as activity level and number of questions can be determined by the user's login account to determine the user's professional situation.

[0039] The text acquisition unit is used to convert the target question into text information. Generally, the target question can be directly entered into the text box to form text information. However, if the user enters a different question interface or uploads an image, the corresponding information may need to be converted to complete the text information. The corresponding method is to extract information elements by identifying the image and extract text data by jumping to the interface through a pre-set button to obtain complete text information. The above information is processed through a pre-set database and calculation strategy to determine the user's professional level, as follows:

[0040] The professional calculation module is associated with a port database, an interaction database, and a professional feature vocabulary, and is configured with an expected calculation strategy. The expected calculation strategy includes

[0041] Step A1. Obtain the corresponding port configuration data from the port database according to the access port data and obtain the corresponding user interaction data from the interaction database according to the user number data; first, there is only static data in the user information, and only the user's port status and user number can be obtained. However, in fact, as the user uses it, the port configuration is different, and the professionalism reflected may also be different. For example, the user will update the corresponding activity every time he uses it, so the information related to the user port is pre-stored in the port database through the port configuration data, and the user number data is pre-existing in the interaction database in the form of user interaction data. After the user passes the verification, the relevant data can be called to evaluate the user's professionalism. On the other hand, by recording or editing the application behavior with later data, the corresponding user interaction data can be updated or the corresponding port configuration data can be modified.

[0042] Step A2: Identify a type of professional terminology in the professional feature vocabulary corresponding to the text information; since the text information is uploaded in the form of fields after acquisition, and the purpose of setting up the professional feature vocabulary is to judge the professional level of the question itself, for example, if the question includes key professional terms, it means that the professional level of this question is relatively high, so by pre-configuring professional terms (higher professional level) in the professional feature vocabulary, the professional level of the question can be judged.

[0043] Step A3: Calculate the expected value of professionalism.

[0044] Where A is the expected value of professionalism, α c1 The port reliability coefficient in the port configuration data reflects the port reliability. For example, the intranet can be divided into different levels, and then the port reliability coefficient can be determined based on the division results. For example, the department or company to which the port belongs can be used as the division basis, and different division levels correspond to different port reliability coefficients. The port reliability coefficient can also be related to the activity level of the port on the platform. c1 The topology reliability coefficient corresponds to the port configuration data. The topology reliability coefficient reflects the topological relationship between ports. For example, the topology reliability coefficient of a port that has a topological relationship with a certain professional port is higher. In this way, multiple ports can be connected to the same gateway, and their corresponding reliability can be obtained through the topology inheritance relationship. The topology reliability coefficient is also assigned. Δa is the preset topology benchmark parameter, and α c2 The user reliability coefficient in the user interaction data reflects the reliability of the information in the user name. For example, information such as position, level, and length of service can reflect the user's professional level, thereby judging the user's professional level. This is also achieved by setting rule assignments. The user's professional level can be judged based on the user's basic information. c2 is the effective response rate corresponding to the user interaction data. The effective response rate is obtained through feedback from the platform interaction data. Each successful response using the intelligent question-answering system will increase the effective response rate. In this way, when retrieving user interaction data, the data can be judged according to the user's usage level. β is the preset frequency weight parameter, which generally ranges from 1.5 to 3. t is the time-weighted interaction frequency in the user interaction data. The time-weighted interaction frequency is the user's activity in using the intelligent question-answering system. If the activity is higher, it means that the user is more familiar with the intelligent question-answering system. By weighting the interaction frequency by time, it is ensured that the interaction frequency of users who do not use the platform for a long time will decay. Δf is the average interaction frequency in the user interaction data. The average interaction frequency will not decay according to the user's usage frequency and can reflect the number of times the user uses it. c3is the correct answer rate in the user interaction data. The correct answer rate reflects the user's historical satisfaction with the answer, and can reflect the accuracy of the user's question from the side. c3 is the preset term reliability coefficient, e n It is the professional weight value corresponding to the nth first-class professional term in the target question, where n is the number of first-class professional terms in the target question. A corresponding professional weight value is set for each first-class professional term, and then when the corresponding first-class professional term appears in the text information, the corresponding professional weight value is superimposed to reflect the professional level of the question. The most important purpose of calculating the professional expectation value is to give different answers to the same question text. For example, in a certain scenario, the voltage in the question text information cannot be determined in terms of semantic recognition to determine what type of voltage value it refers to. For a certain professional field, it may refer to the rated voltage. The terminology in the professional field is accustomed to expressing voltage by the rated voltage. At this time, for professionals, semantic recognition should give the result of querying the rated voltage, but for non-professionals, additional data may be required. Therefore, based on the above purpose, the present invention sets the following content:

[0045] The semantic recognition of the present invention still needs to work in the case of missing text information: the missing analysis module includes a format matching unit and a missing identification unit, the format matching unit is configured with a standard format database, the standard format database is configured with a number of standard format data, the format matching unit matches the corresponding standard format data according to user information and text information, and the missing identification unit identifies missing feature data in the text information according to the standard format data; first, the purpose of the format matching unit is to match the standard format corresponding to the text information as the basis for missing judgment, the missing analysis module includes a part-of-speech extraction unit, the part-of-speech extraction unit extracts the corresponding part-of-speech items according to the knowledge element map, and generates the standard format data according to the part-of-speech items, and the part-of-speech items reflect the requirements for words in the standard format data. The format matching unit is configured with a format matching strategy, and the format matching strategy includes

[0046] Step B1: configure a first matching threshold, calculate the matching result between the text information and each standard format data to obtain a first matching value, select the standard format data whose first matching value is greater than the first matching threshold to generate a first matching data group, and χ1=(∑g k ) / G, where k=1, 2, ... i, χ1 is the first matching value, g kis the part-of-speech weight of the kth part-of-speech item that matches successfully, i is the number of part-of-speech items that match successfully, and G is the sum of the part-of-speech weights of all part-of-speech items of the standard format data; first, since we need to find a format data match that is closest to the text information, we calculate the first matching value, and the matching is performed through the part-of-speech items, and the part-of-speech is refined for the features of the text information, such as place names, plans, responsible parties, voltage, etc. If the match is successful, the pre-configured corresponding part-of-speech weight can be obtained, and the proportion of the sum of the matching weights to the total standard format data indicates the degree of matching between the text information and a standard format data. The matching weight is obtained based on the similarity of the part-of-speech items. After the first matching value is calculated, it can be screened to filter out the standard format data with a smaller first matching value to reduce the amount of calculation.

[0047] Step B2: Match the unmatched part-of-speech items of the standard format data in the first matching group with the user information to obtain a second matching value, and χ2=(∑h k ) / G, where k=1, 2, ... j, χ2 is the second matching value, h k is the part-of-speech weight of the kth successfully matched part-of-speech item, and j is the number of successfully matched part-of-speech items; and then supplementary matching is performed through user information, for example, information that is incomplete in the matching result can be supplemented.

[0048] Step B3: Calculate the weighted matching value of each standard format data in the first matching data group, with x s =d s χ1+χ2, where χ s is the weighted matching value, d s It is a preset matching weight parameter, which takes a value between 0.3 and 0.8. The standard format data with the highest weighted matching value is determined as the matching result. In this way, the closest matching result can be obtained from the standard format data, so as to further determine whether there is text missing. The standard format data is obtained by extracting the part-of-speech features of the knowledge graph. The missing recognition unit obtains the part-of-speech items that were not successfully matched in the standard format data in step B1, and generates missing feature data based on the part-of-speech items and text information. By matching the part-of-speech items, the missing feature data can be identified, thereby providing support for completing the data. The specific semantic recognition module is designed as follows:

[0049] The semantic support module is configured with a first support unit, a second support unit, and a third support unit. The semantic support module calls the first support unit, the second support unit, or the third support unit according to whether the expected professional value falls within a preset benchmark expectation range; three benchmark expectation ranges are set, corresponding to the first level, the second level, and the third level. If the benchmark expectation range is at the first level, it means that the user's professional level is judged to be low, and then the first support unit is used for adaptation; if the benchmark expectation range is at the second level, it means that the user's professional level is judged to be medium, and then the second support unit is used for adaptation; if the benchmark expectation range is at the third level, it means that the user's professional level is judged to be high, and then the third support unit is used for adaptation. Preferably, the third level can cover the functions of the second level, and the second level can cover the functions of the first level. For example, if the third response module corresponding to the third support unit cannot obtain all the supplementary information, it can be tried through the second response module. If the second response module corresponding to the second support unit cannot obtain all the supplementary information, it can be tried through the first response module.

[0050] The first support unit outputs a first missing instruction to the application layer based on the missing feature data. The application layer is configured with a first response module. The first response module generates missing selection information based on the first missing instruction and outputs it to await user input of supplementary information. The first support unit extracts a missing feature index from the text information. The first missing instruction is generated based on the missing feature index. The first response module is associated with a selection index database, which pre-stores a number of missing selection information. The missing selection information is retrieved based on the missing feature index. For example, if a user asks about the average household capacity data in Fangshan District, Beijing, in February 2022, and the noun portion of the capacity data is recognized as valid data information, the missing selection information obtained may be: comprehensive data, urban network data, or rural network data. This missing selection information can then be selected by the user. The first response module generates missing selection information based on the first missing instruction and then outputs "Excuse me, are you asking about comprehensive data, urban network data, or rural network data?" for the user to select. After the user completes their selection, the corresponding supplementary information is obtained.

[0051] The second support unit outputs a second missing instruction to the knowledge layer based on the missing feature data. The knowledge layer is configured with a second response module. The second response module is configured with a knowledge format library. The knowledge format library stores a number of graph index formats. The second response module matches the graph index format according to the second missing instruction and generates supplementary information based on the graph index format. The second response module is also configured with a standard text database. The standard text database stores a number of standard text data. The corresponding standard text data is retrieved according to the obtained graph index format, and the difference data between the standard text data and the text information is obtained to generate the supplementary information. The second response module also includes a first coordination unit. The first coordination unit is configured to generate missing selection information based on the difference data and output it to the application layer when multiple difference data are generated at the same time. The purpose of the second missing instruction is to extract the content of the knowledge graph to achieve data completion. For example, the question is: What are the policies on the electricity price formation mechanism issued by the National Development and Reform Commission this year? The missing data after identification is the content included in the electricity price formation mechanism, and the corresponding files can be indexed through the knowledge graph. The keywords of the files can be extracted to complete the corresponding keywords, which are new energy transformation and coal. As supplementary information, they can be corresponded to policy documents in semantic recognition. The documents No. 206 of the National Development and Reform Commission on February 10, 2022 and No. 78 of the State Energy Development and Supervision Commission on February 25, 2022 can complete the identification. For example, when pointing to multiple questions, the question is which policy on the electricity price formation mechanism was issued by the National Development and Reform Commission this year, then the corresponding information can be output to him for selection to generate supplementary information.

[0052] The third support unit outputs the third missing instruction to the data layer based on the missing feature data; the data layer is configured with a third response module, the third response module is configured with an industry terminology database, the industry terminology database is configured with several industry terms, and each industry term is configured with associated index information, the third response module retrieves the corresponding industry terminology based on the matching of the third missing instruction and the associated index information, and generates supplementary information. For users with a high degree of professionalism, they may ask questions using general terms or simple terms, especially older scholars. This kind of situation will originally make problem solving more difficult and make it impossible to achieve industry identification. The third response module is set up by analyzing the industry terminology corresponding to the missing information. For example, if the missing feature index is voltage, then it is necessary to determine a supplementary information from expressions such as rated voltage, reference voltage, peak voltage, etc. The determination method is as follows: the third support unit includes extracting the missing feature index from the text information, the third missing instruction is generated according to the missing feature index, and the supplementary priority value of each industry term related to the missing feature index is calculated according to the associated index information based on the missing feature index. Where k = 1, 2, ... x, Q is the supplementary priority value, δ1 is the preset frequency priority parameter, q1 is the first frequency in the associated index information, which reflects the total number of times the industry term appears in different databases of the data layer, u x The database priority corresponding to the x-th database in the data layer. Since different databases correspond to different industry specialties and different degrees of commonality, the vocabulary priority is also configured for different databases. t0 is the current time, t x is the last update time corresponding to the x-th database. As time changes, industry terms will also change. q2 is the second frequency in the associated index information. δ2 is the preset associated priority parameter. The second frequency reflects the number of times the industry term is used in the application layer. δ3 is the baseline priority parameter in the associated index information. The third response module selects a supplementary priority value that meets the first preset condition to generate the supplementary information. The first preset condition is configured with a first recognition threshold and a second recognition threshold. When the supplementary priority value corresponding to the industry term is greater than the first supplementary threshold, or the supplementary priority value corresponding to the industry term is greater than the second recognition threshold of the average priority value, the first recognition condition is considered to be met. The average priority value is the average of the supplementary priority values ​​of all industry terms.

[0053] The third response module includes a second coordination unit, which is configured to generate missing selection information based on the industry terms and output it to the application layer when multiple industry terms that meet the first preset condition exist. Similarly, when multiple industry terms that meet the condition exist, the user can be allowed to select one.

[0054] Of course, the above are only typical examples of the present invention. In addition, the present invention may also have many other specific implementation methods. Any technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection required by the present invention.

Claims

1. A power grid intelligent question-answering system based on question information missing completion, comprising an application layer, a support layer, a knowledge layer, and a data layer, characterized by: The support layer is configured with a professionalism calculation module, a missing analysis module, and a semantic support module. The professionalism calculation module calculates the expected professionalism value of the target question based on user information and text information; The missing analysis module includes a format matching unit and a missing identification unit. The format matching unit is configured with a standard format database, which is configured with a number of standard format data. The format matching unit matches the corresponding standard format data according to the user information and the text information. The missing identification unit identifies missing feature data in the text information based on the standard format data. The semantic support module is configured with a first support unit, a second support unit, and a third support unit. The semantic support module calls the first support unit, the second support unit, or the third support unit according to whether the professional expectation value falls within a preset benchmark expectation range; the first support unit outputs a first missing instruction to the application layer according to the missing feature data, the second support unit outputs a second missing instruction to the knowledge layer according to the missing feature data, and the third support unit outputs a third missing instruction to the data layer according to the missing feature data; The application layer is configured with a first response module, which generates missing selection information according to the first missing instruction and outputs it to wait for the user to input supplementary information; The knowledge layer is configured with a second response module, the second response module is configured with a knowledge format library, the knowledge format library stores a plurality of graph index formats, the second response module matches the graph index format according to the second missing instruction, and generates supplementary information according to the graph index format; The data layer is configured with a third response module, and the third response module is configured with an industry terminology database. The industry terminology database is configured with a number of industry terms, and each industry term is configured with associated index information. The third response module retrieves the corresponding industry term according to the matching of the third missing instruction and the associated index information to generate supplementary information.

2. The power grid intelligent question-answering system based on question information missing completion according to claim 1, characterized in that: The professionalism calculation module includes an information acquisition unit and a text acquisition unit; The information acquisition unit is used to acquire user information, the user information including access port data and user number data, the access port data reflects the address of the user access terminal, and the user number data reflects the account used by the user; The text acquisition unit is used to convert the target question into text information; The professional calculation module is associated with a port database, an interaction database, and a professional feature vocabulary, and is configured with an expected calculation strategy. The expected calculation strategy includes Step A1: acquiring corresponding port configuration data from a port database according to the access port data and acquiring corresponding user interaction data from an interaction database according to the user number data; Step A2: identifying a type of professional term in a professional feature vocabulary corresponding to the text information; Step A3: Calculate the expected value of professionalism. Where A is the expected value of professionalism, α c1 is the port reliability coefficient in the port configuration data, a c1 is the topology reliability coefficient corresponding to the port configuration data, Δa is the preset topology benchmark parameter, α c2 is the corresponding user reliability coefficient in the user interaction data, a c2 is the corresponding effective response rate in the user interaction data, β is the preset frequency weight parameter, and f t is the time-weighted interaction frequency in the user interaction data, Δf is the average interaction frequency in the user interaction data, and a c3 is the correct response rate in user interaction data, α c3 is the preset term reliability coefficient, e n is the professional weight value corresponding to the nth first-class professional term in the target question, where n is the number of first-class professional terms in the target question.

3. The power grid intelligent question-answering system based on question information missing completion according to claim 1, characterized in that: The missing analysis module includes a part-of-speech extraction unit, which extracts corresponding part-of-speech items according to the knowledge element graph and generates the standard format data according to the part-of-speech items. The part-of-speech items reflect the requirements for words in the standard format data.

4. The power grid intelligent question-answering system based on question information missing completion according to claim 2, characterized in that: The format matching unit is configured with a format matching strategy, and the format matching strategy includes Step B1: configure a first matching threshold, calculate the matching result between the text information and each standard format data to obtain a first matching value, select the standard format data whose first matching value is greater than the first matching threshold to generate a first matching data group, and χ1=(∑g k ) / G, where k=1, 2, ... i, χ1 is the first matching value, g k is the part-of-speech weight of the kth part-of-speech item that matches successfully, i is the number of part-of-speech items that match successfully, and G is the sum of the part-of-speech weights of all part-of-speech items in the standard format data; Step B2: Match the unmatched part-of-speech items of the standard format data in the first matching group with the user information to obtain a second matching value, and χ2=(∑h k ) / G, where k=1, 2, ... j, χ2 is the second matching value, h k is the part-of-speech weight of the kth part-of-speech item that matches successfully, and j is the number of part-of-speech items that match successfully; Step B3: Calculate the weighted matching value of each standard format data in the first matching data group, with x s =d s χ1+χ2, where χ s is the weighted matching value, d s The preset matching weight parameter is used to determine the standard format data with the highest weighted matching value as the matching result.

5. The power grid intelligent question-answering system based on question information missing completion according to claim 4, characterized in that: The missing identification unit obtains the part-of-speech items that are not successfully matched in the standard format data in step B1, and generates missing feature data according to the part-of-speech items and text information.

6. The power grid intelligent question-answering system based on question information missing completion according to claim 1, characterized in that: The first supporting unit includes extracting a missing feature index from text information, the first missing instruction is generated based on the missing feature index, the first response module is associated with a selection index database, the selection index database pre-stores a number of missing selection information, and the corresponding missing selection information is retrieved based on the missing feature index.

7. The power grid intelligent question-answering system based on question information missing completion according to claim 1, characterized in that: The second response module is also configured with a standard text database, which stores a number of standard text data. The corresponding standard text data is retrieved according to the obtained atlas index format, and the difference data between the standard text data and the text information is obtained to generate the supplementary information.

8. The power grid intelligent question-answering system based on question information missing completion according to claim 7, characterized in that: The second response module further includes a first coordination unit, which is configured to generate missing selection information according to the difference data and output it to the application layer when a plurality of difference data are generated simultaneously.

9. The power grid intelligent question-answering system based on question information missing completion according to claim 1, characterized in that: The third supporting unit includes extracting a missing feature index from the text information, generating the third missing instruction based on the missing feature index, calculating the supplementary priority value of each industry term related to the missing feature index based on the associated index information, and Where k = 1, 2, ... x, Q is the supplementary priority value, δ1 is the preset frequency priority parameter, q1 is the first frequency in the associated index information, which reflects the total number of times the industry term appears in different databases of the data layer, u x is the database priority corresponding to the x-th database in the data layer, t0 is the current time, t x is the last update time corresponding to the x-th database, q2 is the second frequency in the associated index information, δ2 is a preset associated priority parameter, the second frequency reflects the number of times the industry term is used in the application layer, and δ3 is the baseline priority parameter in the associated index information; the third response module selects the supplementary priority value that meets the first preset condition to generate the supplementary information.

10. The power grid intelligent question-answering system based on question information missing completion according to claim 9, characterized in that: The first preset condition is configured with a first recognition threshold and a second recognition threshold. When the supplementary priority value corresponding to the industry term is greater than the first supplementary threshold, or the supplementary priority value corresponding to the industry term is greater than the second recognition threshold of the average priority value, the first recognition condition is considered to be met. The average priority value is the average of the supplementary priority values ​​of all industry terms. The third response module includes a second coordination unit, which is configured to generate missing selection information based on the industry terms and output it to the application layer when there are multiple industry terms that meet the first preset condition at the same time.

Images