Recommended question determination method and device based on financial question and answer characteristics

Abstract

The invention provides a recommendation question determination method and device based on financial question and answer features, which can be used in the artificial intelligence field, the financial field or other fields, and the method comprises the following steps: carrying out word segmentation processing on stock questions by using a word segmentation system, determining a plurality of candidate words and word frequencies of the candidate words, and determining feature values corresponding to the candidate words; sorting the feature values, determining vector space dimensions, performing similarity calculation on each vector, and determining a similar problem set; performing entity recognition on financial preset knowledge points to obtain a domain knowledge graph, determining similarity between the knowledge points, and determining an effective question set according to a similarity calculation result between the knowledge points; and according to the questions input by the user, screening the similar question set and the effective question set to obtain a recommended effective question set. The problem of how to quickly recommend effective questions to the user in the prior art is solved, the provided questions and answers better conform to the expectation of the user, the recommendation efficiency and accuracy are improved, human resources are saved, and the user experience is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing and can be used in the field of artificial intelligence, especially a method and device for determining recommended questions based on financial question-and-answer features. Background Art

[0002] The calculation of Chinese word semantic similarity can be applied to many research fields such as natural language processing, intelligent retrieval, and automatic answering. Currently, there are mainly two methods for similarity calculation: the statistics-based method and the semantic dictionary-based method. The statistics-based method is mainly based on domain corpora, and the semantic dictionary-based method is mainly based on the calculated values of recorded words. For financial question-and-answer applications, it is currently impossible to use the existing corpora of hundreds of thousands of questions and answers in the industry for similarity calculation and combine question-and-answer feature values to intervene in search results to make the search accuracy more consistent with the actual search results of users. Therefore, the existing technologies have problems such as low efficiency and poor accuracy in question-and-answer recommendations. Summary of the Invention

[0003] Aiming at the problems existing in the prior art, the main purpose of the embodiments of the present invention is to provide a method and device for determining recommended questions based on financial question-and-answer features, so as to improve the efficiency and accuracy of question recommendation.

[0004] To achieve the above purpose, the embodiments of the present invention provide a method for determining recommended questions based on financial question-and-answer features. The method includes:

[0005] Using a pre-established word segmentation system to perform word segmentation on a preset stock of questions, determining multiple candidate words and the word frequency of each candidate word, and determining the feature value corresponding to each candidate word according to the word frequency of each candidate word;

[0006] Sorting the feature values corresponding to each candidate word to determine the vector space dimension, and using a similarity algorithm to calculate the similarity of each vector in the vector space dimension, and determining a set of similar questions according to the vector similarity calculation result;

[0007] Performing entity recognition on preset financial preset knowledge points to obtain a domain knowledge graph, and using the domain knowledge graph and a similarity algorithm to determine the similarity between knowledge points, and determining a set of valid questions according to the knowledge point similarity calculation result;

[0008] According to the obtained user input question, screening the set of similar questions and the set of valid questions to obtain a set of recommended valid questions; wherein, the set of recommended valid questions is used for question recommendation according to the question information input by the user.

[0009] Optionally, in an embodiment of the present invention, performing word segmentation on a preset stock of questions to determine multiple candidate words and the word frequency of each candidate word includes:

[0010] Perform word segmentation on the preset inventory problems, and perform word segmentation filtering on the inventory problems after word segmentation to determine multiple candidate words;

[0011] Count the number of occurrences of each candidate word to obtain the word frequency of each candidate word.

[0012] Optionally, in an embodiment of the present invention, sort the eigenvalue corresponding to each candidate word, and determine that the vector space dimension includes:

[0013] Sort the eigenvalue corresponding to each candidate word, and select a preset number of candidate words from large to small as the vector space dimension.

[0014] Optionally, in an embodiment of the present invention, according to the obtained user input problem, screen the similar problem set and the effective problem set to obtain a recommended effective problem set, including:

[0015] Obtain the first problem input by the user, extract the eigenvalue of the first problem, and use the eigenvalue of the first problem to match in the similar problem set and the effective problem set to determine the problem set to be selected;

[0016] Obtain the second problem input by the user, use the similarity algorithm to calculate the similarity between the first problem and the second problem, and determine the click effective value corresponding to each problem in the problem set to be selected according to the problem similarity calculation result;

[0017] Obtain the recommended effective problem set according to the click effective value.

[0018] Optionally, in an embodiment of the present invention, determining the click effective value corresponding to each problem in the problem set to be selected according to the problem similarity calculation result includes:

[0019] Compare the problem similarity calculation result with a preset similarity threshold to obtain a similarity comparison result;

[0020] Determine the click effective value corresponding to each problem in the problem set to be selected according to the similarity comparison result.

[0021] Optionally, in an embodiment of the present invention, obtaining the recommended effective problem set according to the click effective value includes:

[0022] Obtain the recommended effective problem set according to the preset effective value threshold and the click effective value, and determine the eigenvalue corresponding to each problem in the recommended effective problem set to obtain the problem key value set corresponding to the recommended effective problem set.

[0023] Optionally, in an embodiment of the present invention, the method further includes:

[0024] Obtain the question information input by the user, and perform word segmentation and eigenvalue extraction processing on the question information to obtain the question information key value;

[0025] Match the question information key value in the question key value set corresponding to the recommended valid question set to determine the recommended questions for the feedback user.

[0026] An embodiment of the present invention also provides a device for determining recommended questions based on financial question and answer features. The device includes:

[0027] A candidate word module, configured to use a pre-established word segmentation system to perform word segmentation on a preset stock of questions, determine multiple candidate words and the word frequency of each candidate word, and determine the feature value corresponding to each candidate word according to the word frequency of each candidate word;

[0028] A similar question set module, configured to sort the feature values corresponding to each candidate word, determine the vector space dimension, and use a similarity algorithm to calculate the similarity of each vector in the vector space dimension, and determine a similar question set according to the vector similarity calculation result;

[0029] An effective question set module, configured to perform entity recognition on preset financial pre-established knowledge points to obtain a domain knowledge graph, and use the domain knowledge graph and a similarity algorithm to determine the similarity between knowledge points, and determine an effective question set according to the knowledge point similarity calculation result;

[0030] A recommended effective question module, configured to screen the similar question set and the effective question set according to the obtained user input question to obtain a recommended effective question set; wherein, the recommended effective question set is used to recommend questions according to the question information input by the user.

[0031] Optionally, in an embodiment of the present invention, the candidate word module includes:

[0032] A candidate word unit, configured to perform word segmentation on a preset stock of questions, and perform word segmentation filtering on the segmented stock of questions to determine multiple candidate words;

[0033] A word frequency unit, configured to count the number of occurrences of each candidate word to obtain the word frequency of each candidate word.

[0034] Optionally, in an embodiment of the present invention, the similar question set module is further configured to sort the feature values corresponding to each candidate word, and select a preset number of candidate words from largest to smallest as the vector space dimension.

[0035] Optionally, in an embodiment of the present invention, the recommended effective question module includes:

[0036] A candidate question unit, configured to obtain a first question input by a user, extract the feature value of the first question, and use the feature value of the first question to perform matching in the similar question set and the effective question set to determine a candidate question set;

[0037] The click valid value unit is used to obtain the second question input by the user, calculate the similarity between the first question and the second question by using the similarity algorithm, and determine the click valid value corresponding to each question in the set of candidate questions according to the calculation result of the question similarity;

[0038] The recommended valid question unit is used to obtain the set of recommended valid questions according to the click valid value.

[0039] Optionally, in an embodiment of the present invention, the click valid value unit includes:

[0040] The comparison result sub-unit is used to compare the calculation result of the question similarity with a preset similarity threshold to obtain a similarity comparison result;

[0041] The click valid value sub-unit is used to determine the click valid value corresponding to each question in the set of candidate questions according to the similarity comparison result.

[0042] Optionally, in an embodiment of the present invention, the recommended valid question unit is further used to obtain the set of recommended valid questions according to a preset valid value threshold and the click valid value, and determine the feature value corresponding to each question in the set of recommended valid questions to obtain the question key value set corresponding to the set of recommended valid questions.

[0043] Optionally, in an embodiment of the present invention, the device further includes:

[0044] The information key value module is used to obtain the question information input by the user, and perform word segmentation and feature value extraction processing on the question information to obtain the question information key value;

[0045] The recommended question module is used to match the question information key value in the question key value set corresponding to the set of recommended valid questions to determine the recommended question to be fed back to the user.

[0046] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above method is implemented.

[0047] The present invention also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program for the computer to execute the above method.

[0048] The present invention also provides a computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, the steps of the above method are implemented.

[0049] The present invention combines a similarity calculation algorithm with a corpus to obtain a recommended effective question set, solves the problem of how to quickly recommend effective questions to users in the prior art, makes the questions and answers provided to users more in line with the expectations of users, improves the recommendation efficiency while enhancing the accuracy of recommendations, saves human resources, and improves the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] 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 the description of the embodiments. Obviously, the drawings in the following description 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.

[0051] Figure 1 It is a flowchart of a method for determining recommended questions based on financial Q&A features in an embodiment of the present invention;

[0052] Figure 2 It is a flowchart of determining candidate words and word frequencies in an embodiment of the present invention;

[0053] Figure 3 It is a flowchart of obtaining a recommended effective question set in an embodiment of the present invention;

[0054] Figure 4 It is a flowchart of determining click effective values in an embodiment of the present invention;

[0055] Figure 5 It is a flowchart of determining recommended questions in an embodiment of the present invention;

[0056] Figure 6 It is a flowchart of determining recommended questions based on financial Q&A features in a specific embodiment of the present invention;

[0057] Figure 7 It is a schematic structural diagram of a device for determining recommended questions based on financial Q&A features in an embodiment of the present invention;

[0058] Figure 8 It is a schematic structural diagram of a candidate word module in an embodiment of the present invention;

[0059] Figure 9 It is a schematic structural diagram of a recommended effective question module in an embodiment of the present invention;

[0060] Figure 10 It is a schematic structural diagram of a click effective value unit in an embodiment of the present invention;

[0061] Figure 11 It is a schematic structural diagram of a device for determining recommended questions based on financial Q&A features in another embodiment of the present invention;

[0062] Figure 12 The structural schematic diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0063] An embodiment of the present invention provides a method and device for determining a recommended question based on financial question-and-answer features, which can be used in the fields of artificial intelligence, finance, and other fields. It should be noted that the method and device for determining a recommended question based on financial question-and-answer features of the present invention can be used in the fields of artificial intelligence and finance, and can also be used in any field other than the fields of artificial intelligence and finance. The application fields of the method and device for determining a recommended question based on financial question-and-answer features of the present invention are not limited.

[0064] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0065] The information collected in the technical solutions of the present invention is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure, and application of relevant data, etc., all comply with the relevant laws, regulations, and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0066] As Figure 1 shown is a flowchart of a method for determining a recommended question based on financial question-and-answer features according to an embodiment of the present invention. The execution subject of the method for determining a recommended question based on financial question-and-answer features provided by the embodiment of the present invention includes but is not limited to a computer. The present invention combines a similarity calculation algorithm with a corpus to obtain a set of recommended valid questions, solves the problem of how to quickly recommend valid questions to users in the prior art, makes the questions and answers provided to users more in line with the expectations of users, improves the recommendation efficiency while improving the recommendation accuracy, saves human resources, and improves the user experience. The method shown in the figure includes:

[0067] Step S1: Using a pre-established word segmentation system, perform word segmentation on a preset stock of questions to determine multiple candidate words and the word frequency of each candidate word, and determine the feature value corresponding to each candidate word according to the word frequency of each candidate word.

[0068] Among them, using a pre-established or existing word segmentation system to perform word segmentation on the stock problems, multiple candidate words can be obtained. Specifically, a word segmentation system can be established in a conventional manner or an existing word segmentation system can be used. The preset stock problems can be a corpus containing a large number of questions and answers. In addition, the word segmentation of the stock problems can be performed using the conventional operations of the word segmentation system, which will not be elaborated here.

[0069] Furthermore, by counting information such as the occurrence times and occurrence positions of candidate words, the word frequency of candidate words and the eigenvalue of candidate words can be determined. Specifically, the eigenvalue of candidate words is calculated using the comprehensively weighted TF-IDF value, and TF i represents the word frequency of the candidate feature word c i and IDF i represents the IDF value of the candidate word c i . The TF-IDF weighted formula for c i is as follows:

[0070]

[0071] Among them, f i is the number of occurrences of the candidate feature word c i , and F(d) is the sum of the number of occurrences of all words in the document d. To ensure the effectiveness of the formula, when c i is an out-of-vocabulary word, the value of D(ci) is set to 1.

[0072] Step S2: Sort the eigenvalues corresponding to each candidate word, determine the vector space dimension, and use the similarity algorithm to calculate the similarity of each vector in the vector space dimension. According to the vector similarity calculation result, determine the set of similar questions.

[0073] Among them, the eigenvalues corresponding to each candidate word are sorted according to size, and the first m items are selected as the vector space dimension, where m is a preset number, that is, the number of eigenvalues selected.

[0074] Furthermore, the similarity algorithm is used to calculate the similarity of each vector. Specifically, the method of calculating the cosine similarity can be used to calculate the vector similarity result. Thus, it is realized that the first m items are selected as the dimension of the vector space by eigenvalue extraction, and the cosine similarity calculation method is used to calculate the cosine value of the angle between two vectors to measure their similarity. The obtained vector similarity calculation result is compared with the preset vector similarity threshold, and the questions greater than the vector similarity threshold are selected to form the set of similar questions.

[0075] Step S3: Perform entity recognition on the preset financial preset knowledge points to obtain the domain knowledge graph, and use the domain knowledge graph and the similarity algorithm to determine the similarity between knowledge points. According to the knowledge point similarity calculation result, determine the set of effective questions.

[0076] Among them, the pre-set financial knowledge points are the content in the pre-set financial corpus and can be set manually. Specifically, use bert entity recognition to establish the domain hierarchy of knowledge point associations to construct a domain knowledge graph, determine the number of nodes in the domain knowledge graph, and use a similarity algorithm to calculate the similarity between knowledge points.

[0077] Specifically, use a conventional method to determine the number of entity nodes in the domain knowledge graph and the number of nodes where the corresponding entities intersect, and use the attribute-based Jaccard similarity to calculate the similarity between knowledge points to obtain the similarity between knowledge points.

[0078] Furthermore, compare the similarity between knowledge points with the pre-set similarity threshold between knowledge points. If the similarity between knowledge points is greater than the similarity threshold between knowledge points, then select the questions greater than the similarity threshold between knowledge points to form an effective question set.

[0079] Step S4, according to the obtained user input question, screen the similar question set and the effective question set to obtain a recommended effective question set; among them, the recommended effective question set is used to recommend questions according to the question information input by the user.

[0080] Among them, the user inputs their question through a terminal or other device, that is, the user input question, as the first question. Extract the feature value of the user input question, and use the feature value of the user input question to match in the effective question set and the similar question set, and select the feature values that are similar or the same as the feature value from the two question sets. Specifically, it can be compared by comparing the difference between the user input question and the feature values in the two question sets. If the difference is less than the pre-set difference threshold, it can be considered that the matching is successful.

[0081] Furthermore, select the questions with successful matches from the similar questions and the effective question set to form a candidate question set. At this time, obtain the user's input question again as the second question, and use a conventional similarity calculation method to determine the similarity between the first question and the second question.

[0082] Furthermore, compare the similarity of the two questions with the pre-set interval, and determine the click effectiveness value of each question according to the comparison result, thereby realizing recording the user operation behavior characteristics through the feedback of the user's clicks.

[0083] Furthermore, the candidate question set can be further screened through the click effectiveness value. Compare the click effectiveness value corresponding to each question in the candidate question set with the pre-set click effectiveness threshold, and screen out the questions whose click effectiveness value is greater than the click effectiveness threshold, thereby completing the screening of the effective question set and the similar question set. The screened questions form the recommended effective question set. The recommended effective question set is used to quickly and accurately provide recommended questions for users.

[0084] As an embodiment of the present invention, as Figure 2 shown, perform word segmentation on the preset inventory problem, and determine multiple candidate words and the word frequency of each candidate word, including:

[0085] Step S11: Perform word segmentation on the preset inventory problem, and perform word segmentation filtering on the inventory problem after word segmentation to determine multiple candidate words;

[0086] Step S12: Count the number of occurrences of each candidate word to obtain the word frequency of each candidate word.

[0087] Among them, use a pre-established or existing word segmentation system to perform word segmentation on the inventory problem, and use a stop word list and a stop part-of-speech list to filter the word segmentation. Stop words mainly include function words that cannot reflect the theme, such as particles like "de", "di", "de", and words that can only reflect the grammatical structure of a sentence like "raner", "yinci", etc. Stop parts of speech include function words, punctuation marks, etc. Thus, multiple candidate words can be obtained.

[0088] Furthermore, record the position and the number of occurrences of the candidate words in the question and answer, and the word frequency of each candidate word can be accurately determined.

[0089] As an embodiment of the present invention, sort the eigenvalue corresponding to each candidate word to determine the vector space dimension, including: sort the eigenvalue corresponding to each candidate word, and select a preset number of candidate words from largest to smallest as the vector space dimension.

[0090] Among them, sort the eigenvalue corresponding to each candidate word according to size, and select the first m items as the vector space dimension, where m is a preset number, that is, the number of eigenvalues selected.

[0091] As an embodiment of the present invention, as Figure 3 shown, according to the obtained user input question, screen the similar question set and the valid question set to obtain a recommended valid question set, including:

[0092] Step S41: Obtain the first question input by the user, extract the eigenvalue of the first question, and use the eigenvalue of the first question to match in the similar question set and the valid question set to determine the set of candidate questions;

[0093] Step S42: Obtain the second question input by the user, use a similarity algorithm to calculate the similarity between the first question and the second question, and determine the click valid value corresponding to each question in the set of candidate questions according to the calculation result of the question similarity;

[0094] Step S43: Obtain a recommended valid question set according to the click valid value.

[0095] Among them, the user inputs their question through devices such as terminals, that is, the user inputs a question, which is used as the first question. Extract the eigenvalue of the user input question, and use the eigenvalue of the user input question to match in the valid question set and the similar question set. Select the eigenvalues that are close to or the same as the eigenvalue from the two question sets. Specifically, it can be compared by comparing the difference between the user input question and the eigenvalues in the two question sets. If the difference is less than the preset difference threshold, it can be considered that the matching is successful.

[0096] Further, select the questions that match successfully from the similar questions and the valid question set to form a set of candidate questions. At this time, obtain the user's input question again as the second question, and use the conventional similarity calculation method to determine the similarity between the first question and the second question.

[0097] In this embodiment, as Figure 4 shown, according to the question similarity calculation result, determining the click valid values corresponding to each question in the candidate question set includes:

[0098] Step S421, compare the question similarity calculation result with the preset similarity threshold to obtain a similarity comparison result;

[0099] Step S422, according to the similarity comparison result, determine the click valid values corresponding to each question in the candidate question set.

[0100] Among them, compare the similarity of the two questions with the preset interval, and determine the click valid value of each question according to the comparison result. For example, a click is invalid for a similarity greater than or equal to 0.5, a click is valid for a similarity less than 0.5, the click valid value is increased by 1 for a similarity value in the range [0 - 0.5), and the click valid value is decreased by 1 for a similarity in the range [0.5 - 1]. Thus, the user operation behavior characteristics are recorded through the feedback of the user's clicks.

[0101] In this embodiment, according to the click valid value, obtaining the recommended valid question set includes: obtaining the recommended valid question set according to the preset valid value threshold and the click valid value, and determining the eigenvalues corresponding to each question in the recommended valid question set to obtain the question key value set corresponding to the recommended valid question set.

[0102] Among them, the candidate question set can be further screened through the click valid value. Compare the click valid values corresponding to each question in the candidate question set with the preset click valid threshold, and screen out the questions with click valid values greater than the click valid threshold. Thus, the screening of the valid question set and the similar question set is completed, and the screened questions constitute the recommended valid question set.

[0103] Further, similar to the financial preset knowledge points and the existing problems, determine the eigenvalues corresponding to each question in the recommended valid question set, and form a question key value set from the eigenvalues of each question.

[0104] In this embodiment, as Figure 5 shown, the method further includes:

[0105] Step S5, obtaining the question information input by the user, and performing word segmentation and eigenvalue extraction processing on the question information to obtain the key value of the question information;

[0106] Step S6, matching the key value of the question information in the set of question key values corresponding to the recommended effective question set to determine the recommended questions to be fed back to the user.

[0107] Among them, in the process of using the recommended effective question set to recommend questions to the user, the question information input by the user is obtained through devices such as terminals. The same as the processing process of financial preset knowledge points and existing questions, word segmentation and eigenvalue extraction processing are performed on the question information to obtain the key value of the question information.

[0108] Further, the key value of the question information is matched with the set of question key values corresponding to the effective question set. Specifically, it can be compared by comparing the difference between the key value of the question information and the eigenvalues in the set of question key values. If the difference is less than the preset difference threshold, it can be considered that the matching is successful.

[0109] Further, if the matching is successful, the questions in the recommended effective question set corresponding to the eigenvalue in the set of question key values are used as the recommended questions to be fed back to the user. Specifically, the recommended questions can be one or more.

[0110] In a specific embodiment of the present invention, the present invention aims at how to use natural language similarity and question and answer features to intervene in the search results in question and answer applications, and quickly generate question recommendations in the case of a large corpus to improve the retrieval efficiency. Specifically, the present invention solves the problem of how to search more accurately in the existing question and answer corpus of hundreds of thousands of questions and answers in the financial industry, overcomes the deficiency that the existing financial industry question and answer system only presents results through reverse index search and does not use the existing large corpus to obtain the similarity value between word texts to intervene and optimize the search results, and provides a recommendation mechanism for quickly providing effective questions based on financial industry questions and answers.

[0111] In this embodiment, the present invention extracts question and answer feature words and calculates the similarity between questions using the existing 220,000 questions and 710,000 answer corpora of banks to obtain an effective question set, and further improves the effective question set by adding the knowledge points preset in the financial corpus. Combining the answer situation of the questions, the best attributes set by the user for the answers, and the expert attributes of the answering users, the effective question set is returned, and user feedback is collected for the returned question set to obtain the final preset question set. When relevant questions are input, question recommendations can be quickly and accurately made, greatly improving the recommendation efficiency. As Figure 6 shown, the specific process is as follows:

[0112] Step 1: Calculate the similarity of the stock problems to obtain the set of valid stock problems

[0113] Step 11: Extract eigenvalue

[0114] The method of vector space model is used to calculate the text similarity. The selection of feature words, that is, the selection of each dimension vector in the vector space, is the key factor affecting the similarity value. It is the refinement and generalization of the text. Therefore, the extraction process of feature keywords affects the accuracy of the entire text similarity calculation. The present invention adopts a method based on statistical eigenvalues to extract feature words. By calculating the word frequency eigenvalues of words from the current 220,000 questions and 710,000 answer corpora of banks, plus screening through the feature word library and stop word library in the financial field, m feature words are extracted from each question according to the eigenvalue size ranking.

[0115] 1) Determination of candidate words. First, use the existing word segmentation system and process the result set. The stop word list and stop word part-of-speech list are used for word segmentation filtering. Stop words mainly include function words that cannot reflect the theme, such as auxiliary words like "de", "di", "de", and words that can only reflect the sentence grammar structure like "ran er", "yin ci", etc. Stop word parts of speech include function words, punctuation marks, etc.

[0116] 2) Record the positions of candidate words in the Q&A. In order to obtain the position information of each word, it is necessary to determine the way to record the position information and the relative importance of the words in each position when reflecting the theme. When the software scans and counts the word frequency word by word, the position information of each word can be recorded.

[0117] 3) Count the word frequency of each candidate word and record information such as part of speech and position.

[0118] 4) Calculate the eigenvalue of each candidate feature word. Use the comprehensive weighted TF-IDF value to calculate the eigenvalue of the candidate word, as shown in formula (1) specifically, and sort the eigenvalues according to the size, and select the top m items as the dimensions of the vector space.

[0119] Step 12: Similarity calculation based on statistical method

[0120] The basic idea of the VSM vector space model is to represent the text as a vector form vi((w1, t1), (w2, t2), …(wj, tj)), where wj represents the feature word representing the text feature, and tj represents the frequency of the feature word wj appearing in the text. Calculating the similarity between two texts is simply transformed into the problem of calculating the similarity value of two text vectors. The most commonly used similarity calculation method in text processing is to use cosine to calculate the similarity value. The advantage of cosine calculation is that it is exactly a number between 0 and 1. If the vector spaces are exactly the same, it is 1, and if they are orthogonal, it is 0, which conforms to the interval range of the similarity value definition.

[0121] Select the first m items through eigenvalue extraction as the dimension of the vector space, and use the cosine similarity calculation method to calculate the cosine value of the angle between two vectors to measure their similarity. The calculation formula of cosine similarity is:

[0122] cos(A,B) = A * B / ||A|| * ||B|| (2)

[0123] Where A and B are two vectors, · represents the dot product of vectors, and ||A|| represents the norm of vector A.

[0124] Calculate the similarity between two questions q i ((w1, t1), (w2, t2), …(wm, tm)), q j The similarity calculation formula for ((W1, T1), (W2, T2), …(Wm, Tm)) is as follows:

[0125]

[0126] Complete the similarity calculation of the current stock problems through the above model, set the threshold α, and generate a set of similar problems {Q1} for those greater than the threshold α.

[0127] Step 2: Combine the financial domain knowledge graph to obtain a set of preset effective problems by preprocessing the financial knowledge points on the platform

[0128] 1) For the 70,000 pre-set financial domain knowledge points on the platform, use bert entity recognition to establish a domain hierarchy related to knowledge points to construct a domain knowledge graph;

[0129] 2) For two questions q i 、q j , P represents the number of nodes of the entities corresponding to the two questions in the knowledge graph, which is the number of intersecting nodes of the corresponding entities. Use the Jaccard similarity based on attributes to calculate the similarity between knowledge points. The calculation formula is:

[0130]

[0131] Complete the similarity calculation of the pre-set knowledge points through the above model, set the threshold β, and obtain and generate a set of effective problems {Q2} for the pre-set financial knowledge points for those greater than the threshold β.

[0132] Step 3: System preprocess the stock effective problem set and the pre-set effective problem set

[0133] The user inputs question A. After extracting the eigenvalues, questions in the question set {Q1} and the question set {Q2} that contain more than 3 replies and have a best answer or expert answer are matched to form a question set {Q3} (i.e., the question set to be selected) and recommended to the user. The system records the number of times each question in the question set {Q3} is clicked, and records question B that the user asks again after clicking. Then, the question similarity between question A and question B is calculated. According to the interval of the similarity value, clicks with a similarity greater than or equal to 0.5 are invalid, and clicks with a similarity less than 0.5 are valid. The click valid value is incremented for clicks with a similarity value in the range [0 - 0.5), and the click valid value is decremented for clicks with a similarity in the range [0.5 - 1]. By recording the user's operation behavior characteristics through the feedback of the user's clicks, for questions with the number of clicks greater than the threshold γ and the click valid value greater than the threshold δ, the recommended valid question set {Q4} and the eigenvalues extracted from the questions in {Q4} are returned, and the eigenvalue key values {W1, W2, W3, W4...}, that is, the question key value set, are output.

[0134] Step 4: Quickly generate a valid question set after the user inputs a question

[0135] After the user inputs the question information, through word segmentation and eigenvalue extraction, when the eigenvalue key value pair is included in the key value pair set, {Q4} is quickly recommended, which greatly improves the recommendation efficiency and accuracy of valid questions. That is, by performing eigenvalue matching with the question key value set corresponding to the recommended valid question set to determine the recommended questions.

[0136] The present invention combines the similarity calculation algorithm with the financial Q&A corpus for the first time. In this way, it solves the problem of how to quickly recommend valid questions to users in the bank service Q&A system, guides users to complete the question closed-loop through querying, and at the same time, because the present invention combines the Q&A evaluation attributes of existing users in the system, the provided questions and answers are more in line with the user's expectations, improving the recommendation efficiency and the accuracy of the recommendation while.

[0137] Such as Figure 7 As shown in the structural schematic diagram of a recommended question determination device based on financial Q&A features according to an embodiment of the present invention, the device shown in the figure includes:

[0138] The candidate word module 10 is used to perform word segmentation processing on the preset stock questions by using the pre-established word segmentation system, determine multiple candidate words and the word frequency of each candidate word, and determine the eigenvalue corresponding to each candidate word according to the word frequency of each candidate word;

[0139] The similar question set module 20 is used to sort the eigenvalues corresponding to each candidate word, determine the vector space dimension, and use the similarity algorithm to calculate the similarity of each vector in the vector space dimension, and determine the similar question set according to the vector similarity calculation result;

[0140] The valid question set module 30 is used to perform entity recognition on preset financial preset knowledge points to obtain a domain knowledge graph, and use the domain knowledge graph and a similarity algorithm to determine the similarity between knowledge points. According to the calculation result of the similarity between knowledge points, a valid question set is determined;

[0141] The recommended valid question module 40 is used to screen the similar question set and the valid question set according to the obtained user input question to obtain a recommended valid question set; wherein, the recommended valid question set is used to recommend questions according to the question information input by the user.

[0142] As an embodiment of the present invention, as Figure 8 shown, the candidate word module 10 includes:

[0143] The candidate word unit 11 is used to perform word segmentation on the preset stock questions, and perform word segmentation filtering on the stock questions after word segmentation to determine a plurality of candidate words;

[0144] The word frequency unit 12 is used to count the occurrence times of each candidate word to obtain the word frequency of each candidate word.

[0145] As an embodiment of the present invention, the similar question set module is further used to sort the feature values corresponding to each candidate word, and select a preset number of candidate words from largest to smallest as the vector space dimension.

[0146] As an embodiment of the present invention, as Figure 9 shown, the recommended valid question module 40 includes:

[0147] The candidate question unit 41 is used to obtain the first question input by the user, extract the feature value of the first question, and use the feature value of the first question to match in the similar question set and the valid question set to determine a candidate question set;

[0148] The click valid value unit 42 is used to obtain the second question input by the user, use the similarity algorithm to calculate the similarity between the first question and the second question, and determine the click valid value corresponding to each question in the candidate question set according to the calculation result of the question similarity;

[0149] The recommended valid question unit 43 is used to obtain a recommended valid question set according to the click valid value.

[0150] In this embodiment, as Figure 10 shown, the click valid value unit 42 includes:

[0151] The comparison result sub-unit 421 is used to compare the calculation result of the question similarity with a preset similarity threshold to obtain a similarity comparison result;

[0152] Click on the valid value sub-unit 422, which is used to determine the click valid values corresponding to each question in the set of candidate questions according to the similarity comparison result.

[0153] In this embodiment, the recommended valid question unit is further configured to obtain a set of recommended valid questions according to a preset valid value threshold and the click valid values, and determine the feature values corresponding to each question in the set of recommended valid questions, so as to obtain a set of question key values corresponding to the set of recommended valid questions.

[0154] In this embodiment, as Figure 11 shown, the apparatus further includes:

[0155] An information key value module 50, configured to obtain the question information input by the user, and perform word segmentation and feature value extraction processing on the question information to obtain a question information key value;

[0156] A recommended question module 60, configured to match the question information key value in the set of question key values corresponding to the set of recommended valid questions to determine the recommended questions to be fed back to the user.

[0157] Based on the same application concept as the above-mentioned method for determining recommended questions based on financial Q&A features, the present invention further provides the above-mentioned apparatus for determining recommended questions based on financial Q&A features. Since the principle of solving problems by this apparatus for determining recommended questions based on financial Q&A features is similar to that of the method for determining recommended questions based on financial Q&A features, the implementation of this apparatus for determining recommended questions based on financial Q&A features can refer to the implementation of the method for determining recommended questions based on financial Q&A features, and the repeated parts will not be elaborated.

[0158] The present invention combines a similarity calculation algorithm with a corpus to obtain a set of recommended valid questions, solves the problem of how to quickly recommend valid questions to users in the prior art, makes the questions and answers provided to users more in line with the expectations of users, improves the recommendation efficiency while improving the recommendation accuracy, saves human resources, and improves the user experience.

[0159] The present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above method is implemented.

[0160] The present invention further provides a computer program product, including computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the above method are implemented.

[0161] The present invention further provides a computer-readable storage medium, which stores a computer program executed by a computer to implement the above method.

[0162] As Figure 12As shown, the electronic device 600 may further include: a communication module 110, an input unit 120, an audio processor 130, a display 160, and a power supply 170. It should be noted that the electronic device 600 does not necessarily have to include Figure 12 all the components shown therein; in addition, the electronic device 600 may further include Figure 12 components not shown in

[0163] As Figure 12 shown, the central processing unit 100, sometimes also referred to as a controller or operation control, may include a microprocessor or other processor device and / or logic device. The central processing unit 100 receives inputs and controls the operations of the various components of the electronic device 600.

[0164] Among them, the memory 140, for example, may be one or more of a buffer, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory, or other suitable devices. It can store the above information related to failures, and can also store programs for executing relevant information. And the central processing unit 100 can execute the programs stored in the memory 140 to implement information storage or processing, etc.

[0165] The input unit 120 provides inputs to the central processing unit 100. The input unit 120 is, for example, a key or a touch input device. The power supply 170 is used to supply power to the electronic device 600. The display 160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0166] The memory 140 may be a solid-state memory. For example, a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It can also be such a memory that stores information even when power is off, can be selectively erased and has more data. Examples of this memory are sometimes referred to as EPROM, etc. The memory 140 may also be some other type of device. The memory 140 includes a buffer memory 141 (sometimes referred to as a buffer). The memory 140 may include an application / function storage unit 142, which is used to store application programs and function programs or the processes for operating the electronic device 600 through the central processing unit 100.

[0167] The memory 140 may further include a data storage unit 143, which is used to store data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 144 of the memory 140 may include various drivers of the electronic device for communication functions and / or for executing other functions of the electronic device (such as a messaging application, an address book application, etc.).

[0168] The communication module 110 is a transmitter / receiver 110 that transmits and receives signals via the antenna 111. The communication module (transmitter / receiver) 110 is coupled to the central processor 100 to provide input signals and receive output signals, which can be the same as in the case of a conventional mobile communication terminal.

[0169] Based on different communication technologies, multiple communication modules 110, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module, etc., can be provided in the same electronic device. The communication module (transmitter / receiver) 110 is also coupled to the speaker 131 and the microphone 132 via the audio processor 130 to provide an audio output via the speaker 131 and receive an audio input from the microphone 132, thereby implementing the usual telecommunication functions. The audio processor 130 can include any suitable buffers, decoders, amplifiers, etc. Additionally, the audio processor 130 is also coupled to the central processor 100, so that it is possible to record on the local machine through the microphone 132 and play the sound stored on the local machine through the speaker 131.

[0170] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0171] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0172] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one process or multiple processes and / or blocksFigure 1 The functions specified in one or more boxes.

[0173] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the steps of the functions specified in one Figure 1 process or more processes and / or boxes Figure 1 or more boxes.

[0174] In the present invention, specific embodiments are used to illustrate the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for determining recommended questions based on financial Q&A features, characterized in that The method includes: Using a pre-established word segmentation system to perform word segmentation on a preset stock of questions, determining multiple candidate words and the word frequency of each candidate word, and determining the eigenvalue corresponding to each candidate word according to the word frequency of each candidate word; Sorting the eigenvalues corresponding to each candidate word, determining the vector space dimension, and using a similarity algorithm to calculate the similarity of each vector in the vector space dimension, and determining a set of similar questions according to the vector similarity calculation result; Performing entity recognition on a preset financial pre-knowledge point to obtain a domain knowledge graph, and using the domain knowledge graph and the similarity algorithm to determine the similarity between knowledge points, and determining a set of effective questions according to the knowledge point similarity calculation result; According to the obtained user input question, screening the set of similar questions and the set of effective questions to obtain a recommended set of effective questions; wherein, the recommended set of effective questions is used to recommend questions according to the question information input by the user.

2. The method according to claim 1, characterized in that, Performing word segmentation on a preset stock of questions to determine multiple candidate words and the word frequency of each candidate word includes: Performing word segmentation on a preset stock of questions, and performing word segmentation filtering on the segmented stock of questions to determine multiple candidate words; Counting the number of occurrences of each candidate word to obtain the word frequency of each candidate word.

3. The method according to claim 1, wherein Sorting the eigenvalues corresponding to each candidate word to determine the vector space dimension includes: Sorting the eigenvalues corresponding to each candidate word, and selecting a preset number of candidate words from largest to smallest as the vector space dimension.

4. The method according to claim 1, characterized in that According to the obtained user input question, screening the set of similar questions and the set of effective questions to obtain a recommended set of effective questions includes: Obtaining a first question input by the user, extracting the eigenvalue of the first question, and using the eigenvalue of the first question to match in the set of similar questions and the set of effective questions to determine a set of candidate questions; Obtaining a second question input by the user, using a similarity algorithm to calculate the similarity between the first question and the second question, and determining the click effective value corresponding to each question in the set of candidate questions according to the question similarity calculation result; Obtaining the recommended set of effective questions according to the click effective value.

5. The method according to claim 4, wherein Determining the click effective value corresponding to each question in the set of candidate questions according to the question similarity calculation result includes: Comparing the question similarity calculation result with a preset similarity threshold to obtain a similarity comparison result; Determining the click effective value corresponding to each question in the set of candidate questions according to the similarity comparison result.

6. The method according to claim 4, wherein Obtaining the recommended set of effective questions according to the click effective value includes: Obtaining the recommended set of effective questions according to a preset effective value threshold and the click effective value, and determining the eigenvalue corresponding to each question in the recommended set of effective questions to obtain a question key value set corresponding to the recommended set of effective questions.

7. The method according to claim 6, wherein The method further includes: Obtaining the question information input by the user, and performing word segmentation and eigenvalue extraction processing on the question information to obtain a question information key value; Matching the question information key value in the question key value set corresponding to the recommended set of effective questions to determine a recommended question for feedback to the user.

8. A recommendation question determination device based on financial question-and-answer features, characterized in that The device includes: A candidate word module, which is used to segment a preset stock of questions by using a pre-established word segmentation system, determine multiple candidate words and the word frequency of each candidate word, and determine the feature value corresponding to each candidate word according to the word frequency of each candidate word; A similar question set module, which is used to sort the feature values corresponding to each candidate word, determine the vector space dimension, and use a similarity algorithm to calculate the similarity of each vector in the vector space dimension, and determine a similar question set according to the vector similarity calculation result; An effective question set module, which is used to perform entity recognition on preset financial preset knowledge points to obtain a domain knowledge graph, and use the domain knowledge graph and a similarity algorithm to determine the similarity between knowledge points, and determine an effective question set according to the knowledge point similarity calculation result; A recommended effective question module, which is used to screen the similar question set and the effective question set according to the obtained user input question to obtain a recommended effective question set; wherein, the recommended effective question set is used to recommend questions according to the question information input by the user.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program for the computer to execute the method according to any one of claims 1 to 7.

Images