An artificial intelligence system for multi-domain question answering

Abstract

The present invention discloses an artificial intelligence system for multi-domain question answering, belonging to the technical field of multi-domain question answering, and comprising an update analysis module, a supplementary acquisition module, and a trust analysis module; the update analysis module is used to perform real-time update analysis on the knowledge data in the corresponding professional field to determine the supplementary data in the corresponding professional field; the supplementary acquisition module is used to perform real-time data acquisition according to the supplementary data to obtain the supplementary acquisition data in the corresponding professional field; the trust analysis module is used to perform real-time trust analysis on the supplementary acquisition data to obtain the trust level corresponding to the corresponding supplementary acquisition data, and the trust level includes an untrusted level, a trust qualified level, and an accurate trust level; the supplementary acquisition data with the trust level of the untrusted level is eliminated, the potential knowledge data corresponding to the remaining supplementary acquisition data is identified, the potential knowledge data is marked as supplementary knowledge data, and the corresponding trust level label is marked for the supplementary knowledge data; the supplementary knowledge data is stored in the knowledge base.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-domain question answering, and specifically is an artificial intelligence system for multi-domain question answering. Background Art

[0002] Multi-domain question-answering systems are a key branch of artificial intelligence. Their core goal is to provide users with accurate, real-time answers to cross-domain questions by integrating heterogeneous knowledge from multiple sources. With the explosive growth of knowledge, such as a 10% annual increase in academic papers and millions of daily news updates, the lag in updating domain knowledge has become a key bottleneck restricting system performance. For example, if the side effects of new medical drugs or breakthrough research results in science and technology are not promptly incorporated into the knowledge base, the system will output outdated or incorrect answers, seriously undermining user trust.

[0003] Based on this, in order to solve the problem of updating knowledge in multiple fields, the present invention provides an artificial intelligence system for multi-field question answering. Summary of the Invention

[0004] In order to solve the problems existing in the above solutions, the present invention provides an artificial intelligence system for multi-domain question answering.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] An artificial intelligence system for multi-domain question answering, comprising a knowledge base, an update analysis module, a supplementary acquisition module, a trust analysis module, and a question answering module;

[0007] The knowledge base is used to store knowledge data in various professional fields.

[0008] The update analysis module is used to perform real-time update analysis on the knowledge data of the corresponding professional field and determine the supplementary data of the corresponding professional field.

[0009] Furthermore, the knowledge data of the corresponding professional fields are updated and analyzed in real time, including:

[0010] Preset timeliness standards for various professional fields, including qualification standards and qualification rate standards;

[0011] Conduct real-time analysis on corresponding professional fields according to timeliness standards to obtain supplementary analysis fields; conduct real-time analysis on supplementary analysis fields according to timeliness standards to obtain corresponding supplementary data.

[0012] Furthermore, real-time analysis of relevant professional fields is conducted based on timeliness standards, including:

[0013] Set up a retrieval unit, which is used to retrieve in real time the knowledge association data not stored in the knowledge base in the corresponding professional field. The knowledge association data includes retrieved knowledge data and verification result data of the corresponding retrieved knowledge data.

[0014] Filter the knowledge association data according to the verification result data to obtain the domain material data of the professional field.

[0015] Obtain in real time the knowledge update records in the corresponding professional field of the knowledge base; perform calibration analysis on each domain material data according to the knowledge update records and timeliness standards to obtain the judgment result on whether the corresponding professional field meets the timeliness standards.

[0016] Further, performing calibration analysis on each domain material data according to the knowledge update records and timeliness standards includes:

[0017] According to the knowledge update records, identify in real time the update waiting duration corresponding to each domain material data; summarize the update waiting durations of each professional field to form the update duration set of the professional field.

[0018] Establish a domain evaluation model, and the expression of the domain evaluation model is: ;

[0019] In the formula: (U, ST) is the input data, U is the update duration set of the corresponding professional field; ST is the timeliness standard of the corresponding professional field; U→ST means that the corresponding update duration set meets the requirements of the timeliness standard; the output data is the domain evaluation value LP(U, ST), and the domain evaluation value is 1 or 0.

[0020] Perform real-time analysis on the update duration set and timeliness standard of the professional field according to the domain evaluation model to obtain the corresponding domain evaluation value.

[0021] Mark the professional fields with a domain evaluation value of 0 as supplementary analysis fields.

[0022] Further, performing real-time analysis on the supplementary analysis fields according to the timeliness standards includes:

[0023] Perform real-time retrieval on the supplementary analysis fields through the retrieval unit to obtain the knowledge retrieval data of the supplementary analysis fields; set the warning duration according to the timeliness standards.

[0024] Obtain in real time the verification result data and update waiting duration of the knowledge retrieval data, and perform real-time screening on the knowledge retrieval data according to the verification result data.

[0025] Perform real-time analysis on the remaining knowledge retrieval data according to the warning duration and the update waiting duration. When the update waiting duration reaches the warning duration, supplement the knowledge retrieval data into the supplementary data in the supplementary analysis field.

[0026] The supplementary acquisition module is used to perform real-time data acquisition according to the supplementary data to obtain supplementary acquisition data in the corresponding professional field. The supplementary acquisition data includes potential knowledge data and verification association data.

[0027] The trust analysis module is used to perform real-time trust analysis on the supplementary acquisition data to obtain the trust level corresponding to the corresponding supplementary acquisition data. The trust levels include untrusted level, trust qualified level, and accurate trust level.

[0028] Eliminate the supplementary acquisition data with the untrusted level of trust, identify the potential knowledge data corresponding to the remaining supplementary acquisition data, mark the potential knowledge data as supplementary knowledge data, and mark the corresponding trust level label for the supplementary knowledge data; store the supplementary knowledge data in the knowledge base.

[0029] Further, classify and store the supplementary knowledge data stored in the knowledge base according to the trust level. The response priority of the accurate trust level is greater than the response priority of the qualified trust level.

[0030] Further, performing real-time trust analysis on the supplementary acquisition data includes:

[0031] Preset the trust value intervals corresponding to each trust level, and the value range of the trust value is [0, 100];

[0032] Establish a trust value evaluation model, which is used to evaluate the trust value of the corresponding potential knowledge data;

[0033] Real-time identify the verification association data of the supplementary acquisition data, analyze the verification association data through the trust value evaluation model, and obtain the trust value of the potential knowledge data corresponding to the supplementary acquisition data;

[0034] Determine the trust level of the potential knowledge data according to the trust value and the trust value interval of the corresponding trust level.

[0035] Further, perform trust level update and adjustment on the supplementary knowledge data stored in the knowledge base.

[0036] The Q&A module is used for users to ask questions, real-time obtain the question data of the users, generate corresponding response data according to the question data, and display the response data to the users.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] Through the mutual cooperation among the knowledge base, the update analysis module, the supplementary acquisition module, the trust analysis module and the Q&A module, the intelligent update of knowledge data in multiple professional fields is realized, ensuring the accuracy and timeliness of the knowledge data, and then realizing intelligent Q&A in multiple fields; it can effectively cope with the challenges brought by the explosive growth of knowledge, ensure that key knowledge such as new drug side effects in the medical field and breakthrough research results in the scientific and technological field are updated to the knowledge base in a timely manner, avoid outdated or incorrect answers caused by knowledge lag, and significantly improve the accuracy and reliability of cross-field Q&A; the output of accurate and real-time answers can directly enhance the user's trust in the multi-field Q&A system; users do not need to worry about decision-making biases caused by obsolete or incorrect knowledge. Especially in fields such as medicine and technology where the timeliness requirements for information are extremely high, the present invention can provide users with more valuable question answers, thereby optimizing the user interaction experience and enhancing user stickiness. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0040] Figure 1 It is a block diagram of the principle of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0042] As Figure 1 shown, an artificial intelligence system for multi-field Q&A includes a knowledge base, an update analysis module, a trust analysis module and a Q&A module;

[0043] The knowledge base is used to store knowledge data in each professional field.

[0044] The update analysis module is used to perform real-time update analysis on the knowledge data in the corresponding professional field, determine the supplementary data in the corresponding professional field, and the supplementary data is used to indicate the knowledge data that needs to be supplemented in this professional field.

[0045] In one embodiment, performing real-time update analysis on the knowledge data in the corresponding professional field includes:

[0046] Identify the various professional fields of the Q&A system service, and preset the timeliness standards for each professional field. For example, when new knowledge data appears in a certain professional field, the requirement of its timeliness standard is to be stored in the knowledge base within one week, and a compliance rate of 90% is considered to meet the timeliness standard. That is, the timeliness standard includes two criteria, one is the qualification standard, and the other is the qualification rate standard; because different professional fields have different requirements for the timeliness of knowledge Q&A, it is necessary to set the timeliness standard according to actual needs and make dynamic adjustments; specifically, it is set by the platform side.

[0047] Perform real-time analysis on the corresponding professional field according to the timeliness standard, judge whether the knowledge data update method of the corresponding professional field meets the requirements, and mark the professional field that does not meet the requirements as the supplementary analysis field;

[0048] Perform real-time analysis on the supplementary analysis field according to the timeliness standard to obtain the corresponding supplementary data.

[0049] In one embodiment, performing real-time analysis on the corresponding professional field according to the timeliness standard includes:

[0050] Set up a retrieval unit, where the retrieval unit is used to retrieve the latest knowledge association data of the corresponding professional field in real time. The knowledge association data includes the retrieved knowledge data and the verification result data of the retrieved knowledge data. The verification result data is to prove whether it is knowledge data over time, that is, whether the knowledge data is true; because there is a large amount of false data in current networks and other channels, to avoid affecting the Q&A accuracy due to rumor data, that is, the knowledge association data is determined in combination with the subsequent actual verification results and has time persistence, and it can be judged as long as the verification results determined from other subsequent sources are available; it is equivalent to retrieving various possible new knowledge data, marking them as retrieved knowledge data, and taking the retrieved knowledge data as a benchmark to obtain the corresponding verification result data in real time; integrating the retrieved knowledge data with verification result data into knowledge association data; retrieving in real time according to the above method to obtain the knowledge association data of the corresponding professional field.

[0051] Screen the knowledge association data, eliminate the knowledge association data with unqualified verification result data, and mark the remaining knowledge association data as field material data.

[0052] Obtain the update record of the knowledge data in the corresponding professional field in the knowledge base in real time, and mark it as the knowledge update record; perform calibration analysis on each field material data according to the knowledge update record and the timeliness standard to obtain the judgment result of whether the corresponding professional field meets the timeliness standard.

[0053] In one embodiment, performing calibration analysis on each field material data according to the knowledge update record and the timeliness standard includes:

[0054] Determine the update waiting duration corresponding to the material data in each field according to the knowledge update record, that is, the interval duration from retrieving the knowledge data of the determined field material data to its update to the knowledge base; if there is no update or the obvious update waiting duration exceeds the time limit standard, there is no need to wait for it to appear in the knowledge update record, and a non-compliant update waiting duration can be directly output; generally, the duration just exceeding the time limit standard is output.

[0055] Summarize the update waiting durations corresponding to each professional field to form an update duration set.

[0056] Establish a field evaluation model, which is used to analyze the update duration set of the corresponding professional field to determine whether it meets the time limit standard; specifically, calculate the qualification rate corresponding to the update duration set according to the qualification standard of the time limit standard, and then make a judgment according to the qualification rate standard of the time limit standard; the expression of the field evaluation model is: ;

[0057] In the formula: (U, ST) is the input data, U is the update duration set of the corresponding professional field; ST is the time limit standard of the corresponding professional field; U→ST means that the corresponding update duration set meets the requirements of the time limit standard; the output data is the field evaluation value LP(U, ST), and the field evaluation value is 1 or 0.

[0058] According to the field evaluation model, conduct real-time analysis on the update duration set and time limit standard of the corresponding professional field to obtain the corresponding field evaluation value.

[0059] Mark the professional fields with a field evaluation value of 0 as supplementary analysis fields.

[0060] In one embodiment, calibration analysis is performed on the material data in each field according to the knowledge update record and the time limit standard, and analysis can also be performed based on current technologies such as artificial intelligence. For example, a field calibration model is established based on a deep learning algorithm, and a corresponding training set is established through artificial means for training. The training set includes input data and output data. The input data is the knowledge update record, field material data, and time limit standard, and the output data is the verification result, which is used to indicate whether the professional field is a supplementary analysis field; real-time analysis is performed according to the field calibration model after successful training.

[0061] In one embodiment, real-time analysis is performed on the supplementary analysis fields according to the time limit standard, including:

[0062] The retrieval unit in the above embodiments performs real-time retrieval on the supplementary analysis field to obtain corresponding knowledge retrieval data; set a warning duration according to the timeliness standard, that is, from when the update waiting duration of the knowledge retrieval data reaches a certain value to issue a warning, and the warning duration is less than or equal to the duration requirement corresponding to the timeliness standard; specifically set the warning duration according to the needs of the platform party, generally set it less than the duration corresponding to the timeliness standard, if the platform party does not set it, then set the warning duration with the duration corresponding to the timeliness standard; obtain the verification result data and the update waiting duration of the corresponding knowledge retrieval data in real time, if there is no corresponding verification result data currently, then the verification result data is none; perform real-time screening on the knowledge retrieval data according to the verification result data, that is, eliminate the knowledge retrieval data with false or inaccurate verification result data, and also retain the data with no verification result data.

[0063] Perform real-time analysis on the corresponding knowledge retrieval data according to the warning duration and the update waiting duration. When the update waiting duration reaches the warning duration and is not eliminated, supplement the knowledge retrieval data to the supplementary data in the supplementary analysis field.

[0064] In one embodiment, perform real-time analysis on the supplementary analysis field according to the timeliness standard, use the existing analysis method, and under the premise of meeting the timeliness standard, collect the knowledge data to be supplemented in the supplementary analysis field in real time, and integrate to obtain supplementary data.

[0065] The supplementary acquisition module is used to perform real-time data acquisition according to the supplementary data to obtain supplementary acquisition data in the corresponding professional field; that is, perform standardized acquisition according to the acquisition mode, requirements, etc. of the knowledge data; the supplementary acquisition data includes potential knowledge data and verification correlation data; the potential knowledge data is the knowledge data acquired according to the corresponding knowledge retrieval data, but has not been supplemented and stored as knowledge data yet; the verification correlation data is the correlation data that can prove whether it is true and correct acquired in real time for the potential knowledge data; subsequent acquisition can be combined with the verification result data.

[0066] The trust analysis module is used to perform real-time trust analysis on the supplementary acquisition data to obtain the trust level corresponding to the corresponding supplementary acquisition data, and the trust level includes an untrustworthy level, a trust qualified level, and an accurate trust level;

[0067] Eliminate the supplementary acquisition data with the untrustworthy level of the trust level, mark the potential knowledge data corresponding to the remaining supplementary acquisition data as supplementary knowledge data, and mark the corresponding trust level label for the corresponding supplementary knowledge data; store the supplementary knowledge data in the knowledge base.

[0068] In one embodiment, classify and store the supplementary knowledge data stored in the knowledge base according to the trust level, that is, preferentially apply the supplementary knowledge data with the accurate trust level during subsequent responses.

[0069] In one embodiment, real-time trust analysis is performed on supplementary collected data. It can be analyzed based on existing trust analysis methods to determine its trust value, and the corresponding trust level is determined according to the trust value and the preset trust level standard. The trust level standard is the trust value intervals corresponding to three trust levels, which are set by the platform side. For example, 0 - 59 is the untrusted level, 60 - 80 is the qualified trust level, and 81 - 100 is the accurate trust level.

[0070] Exemplarily, a classification model is trained using machine learning algorithms (such as decision trees, random forests, support vector machines, etc.), and the authenticity of the data is predicted according to data features (such as source, content, propagation path, etc.). A large amount of labeled network data (true or false) is collected, features are extracted and the model is trained, and then new data is predicted to output the trust value.

[0071] Deep learning techniques (such as convolutional neural network CNN, recurrent neural network RNN, Transformer, etc.) are used to process multi-modal data such as text, images, and videos to capture complex patterns and features in the data. A deep learning model is constructed, the feature representation of the network data is input, and the authenticity probability or trust value of the data is output.

[0072] NLP techniques are used to analyze the text content of network data, identify information such as semantics, sentiment, and topics, and assist in judging the authenticity of the data. The text is tokenized, part-of-speech tagged, named entity recognized, etc., key information is extracted, and authenticity evaluation is combined with machine learning or deep learning models.

[0073] Combining the prediction results of multiple intelligent models (such as classification models, deep learning models, NLP models, etc.), the final trust value is obtained through methods such as voting and weighted averaging. Multiple base models are trained, and their prediction results are fused to improve the accuracy and stability of the evaluation.

[0074] The intelligent model for evaluating the above trust value is marked as the trust value evaluation model.

[0075] In one embodiment, trust level update analysis is performed on the supplementary knowledge data stored in the knowledge base. The update analysis is performed using the method for determining the trust level to determine whether its trust level changes. At this time, the verification association data can also include data with verification effects such as the Q&A feedback and records of the Q&A personnel.

[0076] In one embodiment, trust level update analysis is performed on the supplementary knowledge data for the qualified trust level.

[0077] In one embodiment, trust level update analysis is performed on all supplementary knowledge data.

[0078] In one embodiment, for the trust level update analysis of supplementary knowledge data, the update analysis can be performed in a timed, regular, interval, real-time manner, etc. For example, different analysis times can be adopted for the qualified trust level and the accurate trust level.

[0079] The Q&A module is used for users to conduct Q&A, obtain the question data of users in real time, answer according to the knowledge base for the question data of users, obtain the corresponding answer data, and display the answer data to users.

[0080] The above formulas are all calculated by removing the dimension and taking their numerical values. The formula is a formula obtained by software simulation of collecting a large amount of data to be closest to the actual situation. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained by simulation of a large amount of data.

[0081] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An artificial intelligence system for multi-domain question answering, comprising a knowledge base and a question answering module. The knowledge base is used to store knowledge data in various professional fields; the question answering module is used for users to ask questions and answer them, to obtain the question data of the user in real time, to generate corresponding response data according to the question data, and to display the response data to the user; characterized in that, It also includes an update analysis module, a supplementary acquisition module, and a trust analysis module; The update analysis module is used to perform real-time update analysis on the knowledge data in the corresponding professional field to determine the supplementary data in the corresponding professional field; The supplementary acquisition module is used to perform real-time data acquisition according to the supplementary data to obtain the supplementary acquisition data in the corresponding professional field, and the supplementary acquisition data includes potential knowledge data and verification association data; The trust analysis module is used to perform real-time trust analysis on the supplementary acquisition data to obtain the trust level corresponding to the corresponding supplementary acquisition data, and the trust level includes an untrusted level, a trust qualified level, and an accurate trust level; Eliminate the supplementary acquisition data with the trust level of untrusted level, identify the potential knowledge data corresponding to the remaining supplementary acquisition data, mark the potential knowledge data as supplementary knowledge data, and mark the corresponding trust level label for the supplementary knowledge data; Store the supplementary knowledge data in the knowledge base; Performing real-time update analysis on the knowledge data in the corresponding professional field includes: Preset the timeliness standards for each professional field, and the timeliness standards include a qualified standard and a pass rate standard; Perform real-time analysis on the corresponding professional field according to the timeliness standards to obtain a supplementary analysis field; perform real-time analysis on the supplementary analysis field according to the timeliness standards to obtain the corresponding supplementary data; Performing real-time analysis on the corresponding professional field according to the timeliness standards includes: Set a retrieval unit, and the retrieval unit is used to retrieve in real time the knowledge association data not stored in the knowledge base in the corresponding professional field, and the knowledge association data includes retrieval knowledge data and verification result data of the corresponding retrieval knowledge data; Filter the knowledge association data according to the verification result data to obtain the field material data of the professional field; Obtain the knowledge update record of the corresponding professional field in the knowledge base in real time; perform calibration analysis on each field material data according to the knowledge update record and the timeliness standards to obtain the judgment result of whether the corresponding professional field meets the timeliness standards.

2. The artificial intelligence system for multi-domain question answering according to claim 1, characterized in that, Performing calibration analysis on each field material data according to the knowledge update record and the timeliness standards includes: According to the knowledge update record, identify the update waiting duration corresponding to each field material data in real time; summarize the update waiting durations of each professional field to form the update duration set of the professional field; Establish a field evaluation model, and the expression of the field evaluation model is: ； In the formula: (U, ST) is the input data, U is the update duration set of the corresponding professional field; ST is the timeliness standard of the corresponding professional field; U→ST means that the corresponding update duration set meets the requirements of the timeliness standard; the output data is the field evaluation value LP(U, ST), and the field evaluation value is 1 or 0; Perform real-time analysis on the update duration set and the timeliness standard of the professional field according to the field evaluation model to obtain the corresponding field evaluation value; Mark the professional field with the field evaluation value of 0 as the supplementary analysis field.

3. The artificial intelligence system for multi-domain question answering according to claim 1, characterized in that, Performing real-time analysis on the supplementary analysis field according to the timeliness standards includes: Perform real-time retrieval on the supplementary analysis field through the retrieval unit to obtain the knowledge retrieval data of the supplementary analysis field; set the warning duration according to the timeliness standards; Obtain the verification result data and the update waiting duration of the knowledge retrieval data in real time, and screen the knowledge retrieval data in real time according to the verification result data; Analyze the remaining knowledge retrieval data in real time according to the warning duration and the update waiting duration. When the update waiting duration reaches the warning duration, supplement the knowledge retrieval data into the supplementary data in the supplementary analysis field.

4. An artificial intelligence system for multi-domain question answering according to claim 1, characterized in that, Classify and store the supplementary knowledge data stored in the knowledge base according to the trust level. The response priority of the answer with the accurate trust level is higher than that of the answer with the qualified trust level.

5. An artificial intelligence system for multi-domain question answering according to claim 1, characterized in that, Conduct real-time trust analysis on the supplementary acquisition data, including: Preset the trust value intervals corresponding to each trust level, and the value range of the trust value is [0, 100]; Establish a trust value evaluation model, which is used to evaluate the trust value of the corresponding potential knowledge data; Identify the verification associated data of the supplementary acquisition data in real time, and analyze the verification associated data through the trust value evaluation model to obtain the trust value of the potential knowledge data corresponding to the supplementary acquisition data; Determine the trust level of the potential knowledge data according to the trust value and the trust value interval of the corresponding trust level.

6. An artificial intelligence system for multi-domain question answering according to claim 1, characterized in that, Update and adjust the trust level of the supplementary knowledge data stored in the knowledge base.

Images