A risk identification method, device and equipment

Abstract

The embodiments of this specification relate to the field of artificial intelligence technology, and specifically disclose a risk identification method, device and equipment, the method comprising: receiving a risk identification request, the risk identification request comprising target risk information of a target object; calculating the number of hit words and the hit word length ratio of the words in the target risk information hitting each risk subject dictionary; wherein the risk subject dictionary comprises at least one risk keyword corresponding to the corresponding risk subject; the hit word length ratio refers to the ratio of the total number of words in the hit words to the total number of words in each word in the target risk information; comparing the number of hit words and the hit word length ratio corresponding to each risk subject dictionary, so as to determine the risk subject to which the target risk information belongs based on the comparison result, thereby improving the accuracy and efficiency of risk identification.

Description

Technical Field

[0001] This specification relates to the field of artificial intelligence technology, and in particular, to a risk identification method, device and equipment. Background Art

[0002] At present, risk prediction for target objects usually adopts manual screening. The traditional manual screening method mainly relies on manual risk identification and verification of events or operation information corresponding to the target object, so as to identify risk events or operations from a large amount of event or operation information. However, the entire screening process requires continuous investment of a large amount of manpower and material resources, and the elimination cycle is long, which is prone to misjudgment. This makes the traditional manual screening method have great limitations and it is difficult to effectively achieve early detection, early disposal and early prevention of risks. Therefore, a more accurate and efficient risk identification method is urgently needed. Summary of the invention

[0003] The purpose of the embodiments of this specification is to provide a risk identification method, device and equipment, which can improve the accuracy and efficiency of risk identification.

[0004] A risk identification method, the method comprising: receiving a risk identification request, the risk identification request comprising target risk information of a target object; calculating the number of hit words and the ratio of hit word lengths of words in the target risk information that hit each risk subject dictionary; wherein the risk subject dictionary comprises at least one risk keyword corresponding to the corresponding risk subject; the hit word length ratio refers to the ratio of the total number of words in the hit words to the total number of words in each word in the target risk information; comparing the number of hit words and the hit word length ratio corresponding to each risk subject dictionary, so as to determine the risk subject to which the target risk information belongs based on the comparison result.

[0005] In some other embodiments of the method provided in the present specification, before calculating the number of hit words and the proportion of hit word length of the words in the target risk information hitting each risk subject dictionary, the method also includes: extracting the words in the target risk information to obtain an initial word set of the target risk information; performing noise word removal processing on the initial word set to obtain a target risk word set of the target risk information; accordingly, calculating the number of hit words and the proportion of hit word length of each word in the target risk word set hitting each risk subject dictionary; the hit word length proportion refers to the ratio of the total number of characters in the hit words to the total number of characters of each word in the target risk word set.

[0006] In some other embodiments of the method provided in this specification, the comparing the number of hit words and the proportion of hit word length corresponding to each risk theme dictionary to determine the risk theme to which the target risk information belongs based on the comparison result includes: arranging the number of hit words corresponding to each risk theme dictionary in descending order; taking the risk theme dictionary ranked first as the designated risk theme dictionary, and taking other risk theme dictionaries as alternative risk theme dictionaries; if the number of hit words of the designated risk theme dictionary is greater than the number of hit words of each alternative risk theme dictionary, then taking the risk theme corresponding to the designated risk theme dictionary as the risk theme to which the target risk information belongs; if there is at least one alternative risk theme dictionary whose number of hit words is equal to the number of hit words of the designated risk theme dictionary, then selecting the risk theme of the risk theme dictionary with the largest proportion of hit word length from the at least one alternative risk theme dictionary and the designated risk theme dictionary as the risk theme to which the target risk information belongs.

[0007] In some other embodiments of the method provided in this specification, the risk theme dictionary is constructed in the following manner: for any specified risk theme, a risk sample document corresponding to the specified risk theme is obtained; a risk word set corresponding to the risk sample document is extracted, and words in the risk word set are used as candidate keywords; the word frequency of each candidate keyword in the risk sample document is calculated; the overall position weight and relative position weight of each candidate keyword in the risk sample document are calculated; wherein the overall position weight is determined according to the ratio of the number of words in the risk sample document to the first word distance of the candidate keyword, and the first word distance is represented by the number of words between the candidate keyword and the first word of the risk sample document; the relative position weight is determined according to the minimum word distance between the candidate keyword and the word representing time in the risk sample document; the inverse document frequency of the candidate keyword is calculated in combination with the overall position weight and the relative position weight; the product of the word frequency and the inverse document frequency is used as a reference value, and candidate keywords whose reference values ​​meet preset requirements are screened out from the risk word set as risk keywords, so as to construct a risk theme dictionary for the specified risk theme based on the risk keywords.

[0008] In some other embodiments of the method provided in this specification, the inverse document frequency of the candidate keyword is calculated by combining the overall position weight and the relative position weight, including:

[0009]

[0010] Among them, W i Risk sample document candidate D j The candidate keywords in W i The inverse document frequency of Dj The number of documents belonging to a given corpus, X i W i The overall position weight, Y i W i The relative position weight, I(W i D j ) in a given document D in the corpus j Contains candidate keywords W i When 1 is selected, it does not contain the candidate keyword W i When 0.

[0011] In some other embodiments of the method provided in this specification, the method also includes: based on the result of verifying the risk keywords in the risk topic dictionary, removing the risk keywords that do not belong to the specified risk topic from the risk topic dictionary and storing them in the interference dictionary of the specified risk topic; extracting the risk word set of other risk sample documents of the specified risk topic based on the interference dictionary to optimize the risk topic dictionary.

[0012] In some other embodiments of the method provided in this specification, the method also includes: obtaining a pre-constructed risk grid; wherein the grid cells of the risk grid correspond to risk themes; pushing the target risk information to the grid cells of the risk themes to which the target risk information belongs, so that the verification subject of the corresponding grid cell verifies whether the risk themes to which the target risk information belongs are correct.

[0013] On the other hand, an embodiment of the present specification also provides a risk identification device, which includes: a receiving module, used to receive a risk identification request, the risk identification request including the target risk information of the target object; a calculation module, used to calculate the number of hit words and the percentage of hit word length in each risk subject dictionary hit by the words in the target risk information; wherein the risk subject dictionary includes at least one risk keyword corresponding to the corresponding risk subject; the hit word length percentage refers to the ratio of the total number of words in the hit words to the total number of words in each word in the target risk information; a comparison module, used to compare the number of hit words and the percentage of hit word length corresponding to each risk subject dictionary; and a subject determination module, used to determine the risk subject to which the target risk information belongs based on the comparison result.

[0014] In some other embodiments of the device provided in this specification, the device also includes a subject dictionary generation module, wherein the subject dictionary generation module includes: an acquisition unit, used for obtaining, for any specified risk topic, a risk sample document corresponding to the specified risk topic; an extraction unit, used for extracting a risk word set corresponding to the risk sample document, and using the words in the risk word set as candidate keywords; a word frequency calculation unit, used for calculating the word frequency of each candidate keyword in the risk sample document; a weight calculation unit, used for calculating the overall position weight and relative position weight of each candidate keyword in the risk sample document; wherein the overall position weight is calculated based on the number of words in the risk sample document. The relative position weight is determined by the ratio of the word frequency to the first word distance of the candidate keyword, wherein the first word distance is represented by the number of words between the candidate keyword and the first word of the risk sample document; the relative position weight is determined according to the minimum word distance between the candidate keyword and the word representing time in the risk sample document; an inverse frequency calculation unit is used to calculate the inverse document frequency of the candidate keyword in combination with the overall position weight and the relative position weight; a screening unit is used to use the product of the word frequency and the inverse document frequency as a reference value, and screen out candidate keywords whose reference values ​​meet preset requirements from the risk word set as risk keywords, so as to construct a risk theme dictionary of the specified risk theme based on the risk keywords.

[0015] On the other hand, an embodiment of the present specification also provides a risk identification device, which includes at least one processor and a memory for storing processor executable instructions, and when the instructions are executed by the processor, the steps of the method described in any one or more of the above embodiments are implemented.

[0016] The risk identification method, apparatus and device provided in one or more embodiments of this specification pre-constructs a risk theme dictionary corresponding to each risk theme, wherein the risk theme dictionary may include at least one risk keyword corresponding to the corresponding risk theme. For the risk information corresponding to any risk event, the number of hit words and the proportion of hit word length of the words in the target risk information hitting each risk theme dictionary can be calculated, and then the number of hit words and the proportion of hit word length corresponding to each risk theme dictionary can be compared to determine the risk theme to which the target risk information belongs based on the comparison result. Using the above method, risk identification can be made more accurate and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0018] Figure 1 A schematic diagram of an implementation flow of a risk identification method in an embodiment provided in this specification;

[0019] Figure 2 A schematic diagram of a process for generating a risk theme dictionary in one embodiment provided in this specification;

[0020] Figure 3 A schematic diagram of a process for risk topic identification in an embodiment provided in this specification;

[0021] Figure 4 A schematic diagram of risk topic identification result verification and risk topic dictionary optimization in one embodiment provided in this specification;

[0022] Figure 5 This is a schematic diagram of the module structure of a risk identification device in an embodiment provided in this specification. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of the specification, not all of the embodiments. Based on one or more embodiments of the specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the embodiments of this specification.

[0024] Figure 1 This is a flow chart of an embodiment of the risk identification method provided in this specification. Figure 1 As shown, the embodiment of this specification also provides a risk identification method, which can be applied to a server that performs risk identification. Accordingly, the method can include the following steps.

[0025] S20: Receive a risk identification request, where the risk identification request includes target risk information of a target object.

[0026] The target object may refer to an object of risk analysis specified by a user, employee, thing, etc. The target risk information may refer to information describing the risk characteristics of the target object. If the target object is an employee, the target risk information may be represented by a document describing a risk event of the employee. For example, a risk event document D is used to describe a risk event of employee A. The content of the risk event document D may be, for example, "Employee A illegally lent out his own account at XX time..." Of course, other forms may also be used to represent the target risk information, which is not limited here.

[0027] The business system may periodically upload a risk identification request of at least one target risk information to the server, and the server may receive the risk identification request. Alternatively, when a business person needs to identify a risk of one or more risk events of a target object, he may also upload the target risk information of the corresponding risk event to the business terminal, and then the business terminal generates a risk identification request and sends it to the server. Currently, the server may also receive the risk identification request in other ways.

[0028] S22: Calculate the number of hit words in each risk theme dictionary and the hit word length ratio of the words in the target risk information; wherein, the risk theme dictionary includes at least one risk keyword corresponding to the corresponding risk theme; the hit word length ratio refers to the ratio of the total number of hit words to the total number of words in each word in the target risk information.

[0029] After receiving the risk identification request, the server can extract the words in the target risk information. If the input format of the content contained in the target risk information is predefined when constructing the target risk information, so that most of the words contained in the target risk information are information describing the risk characteristics of the target object, such as "employee A, XX time, illegal lending, personal account". Then the server can directly extract the words "employee, A, XX, time, illegal, lending, personal, account" in the target risk information.

[0030] If the construction of the target risk information does not set a more strict entry format, and adopts the commonly used risk event description document and other forms. Then the words extracted from the target risk information contain a lot of noise, and the extracted words need to be processed to remove noise words to reduce the impact of noise words on the accuracy of risk identification. In some embodiments, the server can extract words from the target risk information to obtain an initial word set for the target risk information. The initial word set is processed to remove noise words to obtain a target risk word set for the target risk information. Accordingly, the number of hit words and the proportion of hit word length of each word in the target risk word set that hits each risk theme dictionary can be calculated.

[0031] For example, the semantic trunk and related semantic components can be extracted from the risk event document D in combination with dependency syntax. For example, if the risk event document D contains the content "employee A has the problem of "renting or illegally lending out his own account (card)", the HanLP tool can be used to analyze the parts of speech in D, analyze the subject-predicate relationship, verb-object relationship, parallel relationship and other relationships of the word, extract the main morphemes of the sentence, and obtain the initial word set.

[0032] Then, the initial word set can be processed for noise reduction, and the target risk word set can be formed by removing the subjects in the word class (the subjects in risk events are usually names of people or things, which do not contribute to the risk theme), adverbs, punctuation marks, numbers, place names and other characters with little meaning. For example, the target risk word set can include "existence", "rental", "violation", "I", "account card", and "problem". Risk analysis is then performed based on each word in the target risk word set.

[0033] The server may retrieve a risk theme dictionary corresponding to each risk theme, wherein the risk theme dictionary may include at least one risk keyword corresponding to the corresponding risk theme.

[0034] In some embodiments, the risk theme dictionary can be constructed in the following manner:

[0035] S221: For any specified risk topic, obtain a risk sample document corresponding to the specified risk topic.

[0036] S222: extract the risk word set corresponding to the risk sample document, and use the words in the risk word set as candidate keywords. The risk word set can be extracted by referring to the above embodiment, such as extracting the semantic trunk and related semantic components from the risk sample document in combination with dependency syntax, and then performing noise word removal processing to obtain the risk word set of the risk sample document.

[0037] S223: Calculate the word frequency of each candidate keyword in the risk sample document.

[0038] S224: Calculate the overall position weight and relative position weight of each candidate keyword in the risk sample document; wherein the overall position weight is determined according to the ratio of the number of words in the risk sample document to the first word distance of the candidate keyword, and the first word distance is represented by the number of words between the candidate keyword and the first word of the risk sample document; the relative position weight is determined according to the minimum word distance between the candidate keyword and the word representing time in the risk sample document.

[0039] S225: Taking the product of the word frequency and the inverse document frequency as a reference value, selecting candidate keywords whose reference values ​​meet preset requirements from the risk word set as risk keywords, so as to construct a risk topic dictionary of the specified risk topic based on the risk keywords.

[0040] like Figure 2 As shown, the subject identifier of the risk subject can be configured in advance as needed. Then, the risk event description documents belonging to each risk subject can be marked in advance by business personnel as risk sample documents of the corresponding risk subject. It is recommended that the number of risk sample documents corresponding to each risk subject should be no less than 50. For example, the risk sample document "The employee has the problem of renting or illegally lending the employee's account (card)..." and the risk sample document "The employee lent his credit card under his name to his nephew and a false transaction occurred..." can be marked under risk subject #1.

[0041] Then, the semantic trunk and related semantic components can be extracted from the risk sample documents in combination with dependency syntax, and then noise word processing can be performed to obtain the risk word set of the risk sample documents. The words in the risk word set are used as candidate keywords. The improved TF-IDF (Term Frequency / Inverse Document Frequency) algorithm can be used to screen out candidate keywords that meet the preset requirements from the candidate keywords as risk keywords for the corresponding risk topics.

[0042] For the risk sample document D j , we can calculate D j Each candidate keyword W in the risk word set i Frequency of words To utilize Characterize the candidate keywords W i In D j The frequency of occurrence in .

[0043]

[0044] Among them, Count(W i ) represents the candidate keyword W i In D j The number of occurrences in |D J | for D j The number of words in all words.

[0045] Usually, the structure and order of risk event description documents are: event time - event overview - detailed description - event consequences. The event topic is more likely to appear in the event overview. Two adjustment weights (X i , Y i ).

[0046] X i Can be used to characterize candidate keywords W i In D j The overall position weight that appears in X i The overall position of the candidate keywords in the document can be more prominently represented, thereby improving the accuracy of risk keyword extraction. The risk sample document D can be extracted j Number of words |D J |, and extract candidate keywords W i The risk sample document D j The number of words between the first word of i The first word distance K i Afterwards, based on |D J |With K i The ratio of X i . Candidate keyword W i In D j The number of occurrences in is usually greater than 1, which can be summarized as W i In D j The first word distance K corresponding to different positions in i , calculate X i . If W can be calculated i The first word distance K corresponding to different positions i The average D J |with The ratio of X i Alternatively, it can be W i The first word distance K corresponding to different positions i Configure the adjustment factor to calculate the adjustment factor D J |with The ratio of X i , the value of the adjustment factor can be determined by analyzing a certain amount of sample data.

[0047] Y i Can be used to characterize candidate keywords W i With D j The relative position weight of the words representing time in . i The relative position between the candidate keywords of the risk theme and the event time description information can be better characterized, and the accuracy of risk theme word extraction can be improved. The candidate keyword W can be calculated i With D j The minimum word distance H between words representing time in i For a candidate keyword W at any position i, we can get the words representing time and the candidate keywords W i The number of words between m, D j There may be more than one word representing time in D. j The words representing time and the candidate keywords W i The number of words m between them, the minimum value of the number of words m is selected as the candidate keyword W at that position i With D j The minimum word distance H between words representing time in i After that, we can synthesize W i In D j The H corresponding to different positions in i , calculate Y i If W i The corresponding H at different positions i The average value of Y i Alternatively, it can be W i The corresponding H at different positions i Configure the adjustment factor to calculate Y based on the adjustment factor i , the value of the adjustment factor can be determined by analyzing a certain amount of sample data.

[0048] In some embodiments, the inverse document word frequency It can be expressed as:

[0049]

[0050] Among them, W i Risk sample document candidate D j The candidate keywords in W i The inverse document frequency of D j The number of documents belonging to a given corpus, X i W i The overall position weight, Y i W i The relative position weight, I(W i D j ) in a given document D in the corpus j Contains candidate keywords W i When 1 is selected, it does not contain the candidate keyword W i When 0.

[0051] Candidate keyword W i In document D j The closer the position is to the front, the i The larger the candidate keyword W is, the i With D jThe closer the time words are, the better the Y i Correspondingly, the inverse document word frequency calculated in the above manner can further highlight the prevalence of candidate keywords that are close to the words representing time, that is, it further highlights the prevalence of candidate keywords in the event overview, and further improves the accuracy of extracting candidate keywords representing risk topics.

[0052] Then, the value:

[0053]

[0054] The above processing is performed on each risk sample document under the specified risk topic to obtain the candidate keywords of each risk sample document.

[0055] For each candidate keyword under the specified risk theme, Sort them in descending order, and extract the top n candidate keywords as the risk keywords of the specified risk topic. The more frequent and uncommon the candidate keywords in the event overview appear in the current document, the more likely it is that the candidate keyword represents the risk topic of the current document. The above method is used to highlight the candidate keywords of this type. The accuracy of extracting candidate keywords representing risk topics can be further improved. For example, n can be an integer greater than zero. For example, n can be 5. For example, the risk keywords can be stored in a set to obtain a risk topic dictionary of the specified risk topic.

[0056] For example, after extracting risk keywords from the 20 risk sample documents marked under risk topic #1, an initial risk topic dictionary is formed. The top five risk keywords (e.g., “late”, “leave early”, “absenteeism”, “long-term”, and “leave”) ranked from largest to smallest constitute the risk theme dictionary of risk theme #1.

[0057] The word frequency of the risk keyword can also be saved, and the risk keywords can be sorted in order from high to low in terms of word frequency. The risk keywords that are ranked low can be checked by business personnel to confirm whether the risk keyword is the risk keyword of the specified risk theme. For example, the business personnel can confirm whether the risk keyword with a lower word frequency in the risk theme dictionary is the risk keyword of risk theme #1. If it is confirmed that it does not belong, the risk keyword that does not belong can be used as an interference word for the specified risk theme and removed from the risk theme dictionary. The server can store the interference word in the interference dictionary of the specified risk theme. Alternatively, the business personnel can also add some risk keywords to the risk theme dictionary based on experience.

[0058] In the process of optimizing and expanding the risk subject dictionary by using other risk sample documents of the specified risk subject, the server can also extract risk word sets of other risk sample documents of the specified risk subject based on the interference dictionary. For example, after extracting the initial word set of the risk sample document, while processing the initial word set based on the above-mentioned noise word removal processing method, it can also be combined with the interference dictionary of the specified risk subject. If the initial word set contains words in the corresponding interference dictionary, they can be removed from the initial word set to obtain the risk word set of the corresponding risk sample document. The server then expands and optimizes the risk subject dictionary based on the risk word set. Automatically removing words in the interference dictionary during the subsequent risk keyword extraction process can continuously improve the accuracy of the risk subject dictionary.

[0059] Accordingly, in some embodiments, the server may remove risk keywords that do not belong to the specified risk topic from the risk topic dictionary based on the result of checking the risk keywords in the risk topic dictionary, and store them in the interference dictionary of the specified risk topic. The server may extract the risk word set of other risk sample documents of the specified risk topic based on the interference dictionary, so as to optimize the risk topic dictionary based on the risk word set and improve the accuracy of the risk topic dictionary.

[0060] Of course, the server may also construct the risk theme dictionary in other ways, such as directly having business personnel store risk keywords in the risk theme dictionary corresponding to each risk theme based on business experience.

[0061] After the server retrieves the risk theme dictionary corresponding to each risk theme, it can calculate the number of hit words and the percentage of hit word length of the words in the target risk information that hit each risk theme dictionary. Of course, if the target risk word set of the target risk information is improved based on the above embodiment, the number of hit words and the percentage of hit word length of each word in the target risk word set that hits each risk theme dictionary can also be calculated. Correspondingly, the percentage of hit word length refers to the ratio of the total number of words in the hit word to the total number of words in the target risk word set.

[0062] Take the target risk word set as an example. For example, the target risk word set 1 is {"existence", "rental", "violation", "person", "account card", "problem"}. Assume that risk theme dictionary #1 includes risk keywords "rental, account card", risk theme dictionary #2 includes risk keywords "credit card, overdraft", and risk theme dictionary #3 includes risk keywords "business, violation".

[0063] The words in the target risk word set 1 that hit the words in the risk theme dictionary #1 are "rental" and "account card". Accordingly, the number of hit words in the risk theme dictionary #1 is 2, and the hit word length accounts for 5 / 13. The words in the target risk word set 1 that hit the word in the risk theme dictionary #2 are "violation". Accordingly, the number of hit words in the risk theme dictionary #2 is 1, and the hit word length accounts for 2 / 13. The words in the target risk word set 1 do not hit the words in the risk theme dictionary #3.

[0064] S24: Compare the number of hit words and the ratio of hit word lengths corresponding to each risk theme dictionary to determine the risk theme to which the target risk information belongs based on the comparison result.

[0065] Based on step S22, the number of hit words and the percentage of hit word length corresponding to each risk theme dictionary can be calculated. In some embodiments, the server can arrange the number of hit words corresponding to each risk theme dictionary in descending order. The risk theme dictionary ranked first is used as the designated risk theme dictionary, and the other risk theme dictionaries are used as alternative risk theme dictionaries. If the number of hit words of the designated risk theme dictionary is greater than the number of hit words of each alternative risk theme dictionary, the risk theme corresponding to the designated risk theme dictionary is used as the risk theme to which the target risk information belongs. If there is at least one alternative risk theme dictionary whose number of hit words is equal to the number of hit words of the designated risk theme dictionary, the risk theme of the risk theme dictionary with the largest percentage of hit word length is selected from the at least one alternative risk theme dictionary and the designated risk theme dictionary as the risk theme to which the target risk information belongs. Through the above-mentioned comparison method, the selection of the risk theme to which the target risk information belongs can be made more accurate and efficient.

[0066] For example, the number of hit words is compared first, and if the number of hit words corresponding to risk theme dictionary #1 is the largest, the risk theme of risk theme dictionary #1 can be used as the risk theme to which the target risk information belongs.

[0067] Assuming that the number of hit words corresponding to risk theme dictionary #1 and risk theme dictionary #2 is equal, but the hit word length of risk theme dictionary #2 is larger than that of risk theme dictionary #1, the risk theme of risk theme dictionary #2 can be used as the risk theme to which the target risk information belongs.

[0068] like Figure 3 As shown, if each word in the target risk topic does not hit any word in the risk topic dictionary, a result of no hit can be output, and business personnel intervene to mark the risk topic corresponding to the target risk topic.

[0069] Of course, those skilled in the art can also configure other comparison methods as needed. For example, combining a large amount of historical data, setting adjustment factors for the number of hit words and the hit length ratio corresponding to each risk theme dictionary, so as to simultaneously combine the two factors of the number of hit words and the hit length ratio to determine the risk theme to which the target risk information belongs.

[0070] In some other embodiments, the server may also obtain a pre-constructed risk grid, wherein the grid cells of the risk grid correspond to risk topics. The target risk information may be pushed to the grid cells of the risk topics to which the target risk information belongs, so that the verification subject of the corresponding grid cell verifies whether the risk topic to which the target risk information belongs is correct.

[0071] like Figure 4 As shown, a risk grid can be pre-constructed, and the risk grid can include multiple grid units. The grid units of the risk grid can be configured to correspond to the risk topics. Of course, given that risk topics may be added at any time, the number of grid units of the risk grid can be configured to be greater than the risk topics, so that after the new risk topics are added, the risk topics can be directly associated with the idle grid units. A corresponding verification subject can be configured for each grid unit. The verification subject can be, for example, a supervisor.

[0072] Assume that the target risk information is represented by a risk description document of a risk event. After determining the risk topic to which a risk event belongs, the server can push the risk event to the grid unit of the risk topic to which the risk event belongs, and remind the verification subject to perform the verification work, or the server can also regularly remind the verification subject to perform the verification work. If it is confirmed that the risk topic to which the risk event belongs is correct, the server can add the risk event to the information library of the corresponding risk topic based on the confirmation operation of the verification subject. If the risk event does not belong to the determined risk topic, the verification subject can adjust the risk topic to which the risk event belongs, or adjust the risk topic to which the risk event belongs by other means. The server can store the risk event in the information library of the adjusted risk topic based on the adjustment operation. If the risk event is not risky, the risk event can be stored in the reference information library.

[0073] By using the risk grid screening method, professional fields can be distinguished and the accuracy of the risk theme verification to which the risk event belongs can be improved. The grid unit corresponding to each risk theme can be flexibly set up with the screening subject suitable for the risk theme as needed, thereby improving the flexibility of information management.

[0074] The server can expand the risk topic dictionary of the corresponding risk topic based on the newly added risk information in the information base of each risk topic. The interference dictionary can also be updated based on the reference information base to improve the accuracy of risk keyword extraction when expanding the risk topic dictionary of the corresponding risk topic. When the number of words in the risk topic dictionary reaches a certain level, words with low word frequency can be eliminated, making the risk keywords in the risk topic dictionary more representative. When a new risk topic needs to be added, the risk topic dictionary of the new risk topic can be constructed and expanded according to the above method.

[0075] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. For details, please refer to the description of the above-mentioned related processing related embodiments, and no further description is given here.

[0076] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0077] like Figure 5 As shown, based on the method provided in the above embodiment, this specification also provides a risk identification device, which is applied to the server, and the device may include:

[0078] The receiving module 102 may be configured to receive a risk identification request, wherein the risk identification request includes target risk information of a target object.

[0079] The calculation module 104 can be used to calculate the number of hit words in the target risk information that hit each risk topic dictionary and the ratio of the hit word length; wherein the risk topic dictionary includes at least one risk keyword corresponding to the corresponding risk topic; the hit word length ratio refers to the ratio of the total number of hit words to the total number of words in each word in the target risk information.

[0080] The comparison module 106 may be used to compare the number of hit words and the percentage of hit word length corresponding to each risk theme dictionary.

[0081] The topic determination module 108 may be configured to determine the risk topic to which the target risk information belongs based on the comparison result.

[0082] In some other embodiments, the apparatus may further include a subject dictionary generating module 110, wherein the subject dictionary generating module 110 may include:

[0083] The acquisition unit may be used to acquire, for any specified risk topic, a risk sample document corresponding to the specified risk topic;

[0084] An extraction unit may be used to extract a risk word set corresponding to the risk sample document, and use the words in the risk word set as candidate keywords;

[0085] A word frequency calculation unit, which can be used to calculate the word frequency of each candidate keyword in the risk sample document;

[0086] The weight calculation unit can be used to calculate the overall position weight and relative position weight of each candidate keyword in the risk sample document; wherein the overall position weight is determined according to the ratio of the number of words in the risk sample document to the first word distance of the candidate keyword, and the first word distance is represented by the number of words between the candidate keyword and the first word of the risk sample document; the relative position weight is determined according to the minimum word distance between the candidate keyword and the word representing time in the risk sample document;

[0087] An inverse frequency calculation unit, which can be used to calculate the inverse document frequency of the candidate keyword in combination with the overall position weight and the relative position weight;

[0088] The screening unit can be used to use the product of the word frequency and the inverse document frequency as a reference value, and screen out candidate keywords whose reference values ​​meet preset requirements from the risk word set as risk keywords, so as to construct a risk topic dictionary of the specified risk topic based on the risk keywords.

[0089] It should be noted that the above-mentioned device may also include other implementations according to the description of the above-mentioned embodiment. The specific implementation methods may refer to the description of the relevant method embodiments, which will not be described one by one here.

[0090] This specification also provides a risk identification device, which may include at least one processor and a memory for storing processor executable instructions, and the instructions, when executed by the processor, implement the steps of the method described in any one or more of the above embodiments. The memory may include a physical device for storing information, which is usually to digitize the information and then store it in a medium using electrical, magnetic or optical methods. The storage medium may include: devices that store information using electrical energy, such as various memories, such as RAM, ROM, etc.; devices that store information using magnetic energy, such as hard disks, floppy disks, magnetic tapes, magnetic core memories, bubble memories, and U disks; devices that store information optically, such as CDs or DVDs. Of course, there are other types of readable storage media, such as quantum memories, graphene memories, and so on.

[0091] It should be noted that the embodiments of this specification are not limited to the situations that must comply with the standard data model / template or the embodiments of this specification. Certain industry standards or slightly modified implementation plans based on the implementation described in the custom method or the embodiment can also achieve the same, equivalent or similar, or predictable implementation effects after deformation of the above-mentioned embodiments. The embodiments obtained by using these modified or deformed data acquisition, storage, judgment, processing methods, etc. can still fall within the scope of the optional implementation plans of this specification.

[0092] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. In the description of this specification, the description of the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of this specification. In this specification, the schematic representation of the above terms does not necessarily target the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, in the absence of contradiction, a person skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0093] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A risk identification method, characterized in that: The method comprises: receiving a risk identification request, wherein the risk identification request includes target risk information of a target object; Calculate the number of hit words and the length ratio of the hit words in each risk theme dictionary of the words in the target risk information; wherein the risk theme dictionary includes at least one risk keyword corresponding to the corresponding risk theme; the hit word length ratio refers to the ratio of the total number of words in the hit words to the total number of words in each word in the target risk information; Compare the number of hit words and the ratio of hit word lengths corresponding to each risk theme dictionary to determine the risk theme to which the target risk information belongs based on the comparison result; Among them, the comparing the number of hit words and the proportion of hit word length corresponding to each risk theme dictionary to determine the risk theme to which the target risk information belongs based on the comparison result includes: arranging the number of hit words corresponding to each risk theme dictionary in descending order; taking the risk theme dictionary ranked first as the designated risk theme dictionary, and taking other risk theme dictionaries as alternative risk theme dictionaries; if the number of hit words of the designated risk theme dictionary is greater than the number of hit words of each alternative risk theme dictionary, then taking the risk theme corresponding to the designated risk theme dictionary as the risk theme to which the target risk information belongs; if there is at least one alternative risk theme dictionary whose number of hit words is equal to the number of hit words of the designated risk theme dictionary, then selecting the risk theme of the risk theme dictionary with the largest proportion of hit word length from the at least one alternative risk theme dictionary and the designated risk theme dictionary as the risk theme to which the target risk information belongs.

2. The method according to claim 1, characterized in that Before calculating the number of hit words and the proportion of hit word lengths in the target risk information hits in each risk theme dictionary, the method further includes: Extracting words from the target risk information to obtain an initial word set of the target risk information; Performing noise word elimination processing on the initial word set to obtain a target risk word set of the target risk information; Correspondingly, the number of hit words and the hit word length ratio of each word in the target risk word set that hits each risk theme dictionary are calculated; the hit word length ratio refers to the ratio of the total number of characters in the hit words to the total number of characters in each word in the target risk word set.

3. The method according to claim 1, characterized in that The risk theme dictionary is constructed in the following way: For any specified risk topic, obtain a risk sample document corresponding to the specified risk topic; Extracting a risk word set corresponding to the risk sample document, and using the words in the risk word set as candidate keywords; Calculate the word frequency of each candidate keyword in the risk sample document; Calculate the overall position weight and relative position weight of each candidate keyword in the risk sample document; wherein the overall position weight is determined according to the ratio of the number of words in the risk sample document to the first word distance of the candidate keyword, the first word distance is represented by the number of words between the candidate keyword and the first word of the risk sample document; the relative position weight is determined according to the minimum word distance between the candidate keyword and the word representing time in the risk sample document; Calculating the inverse document frequency of the candidate keyword by combining the overall position weight and the relative position weight; Taking the product of the word frequency and the inverse document frequency as a reference value, candidate keywords whose reference values ​​meet preset requirements are screened out from the risk word set as risk keywords, so as to construct a risk topic dictionary of the specified risk topic based on the risk keywords.

4. The method according to claim 3, characterized in that The calculating the inverse document frequency of the candidate keyword in combination with the overall position weight and the relative position weight includes: Among them, W i is the risk sample document D j The candidate keywords in W i The inverse document frequency of D j The number of documents belonging to a given corpus, X i W i The overall position weight, Y i W i The relative position weight, I(W i D j ) in a given document D in the corpus j Contains candidate keywords W i When 1 is selected, it does not contain the candidate keyword W i When 0.

5. The method according to claim 3, characterized in that: The method further comprises: Based on the result of checking the risk keywords in the risk theme dictionary, the risk keywords that do not belong to the specified risk theme are removed from the risk theme dictionary and stored in the interference dictionary of the specified risk theme; A risk word set of other risk sample documents of the specified risk theme is extracted based on the interference dictionary to optimize the risk theme dictionary.

6. The method according to claim 1, characterized in that The method further comprises: Obtaining a pre-constructed risk grid; wherein the grid cells of the risk grid correspond to risk themes; The target risk information is pushed to the grid unit of the risk theme to which the target risk information belongs, so that the verification subject of the corresponding grid unit verifies whether the risk theme to which the target risk information belongs is correct.

7. A risk identification device, characterized in that: The device comprises: A receiving module, configured to receive a risk identification request, wherein the risk identification request includes target risk information of a target object; A calculation module, used to calculate the number of hit words in the target risk information that hit each risk theme dictionary and the hit word length ratio; wherein the risk theme dictionary includes at least one risk keyword corresponding to the corresponding risk theme; the hit word length ratio refers to the ratio of the total number of words in the hit words to the total number of words in each word in the target risk information; A comparison module, used to compare the number of hit words and the percentage of hit word length corresponding to each risk theme dictionary; A theme determination module, used to determine the risk theme to which the target risk information belongs based on the comparison result; The subject dictionary generation module includes: an acquisition unit, for acquiring, for any specified risk subject, a risk sample document corresponding to the specified risk subject; an extraction unit, for extracting a risk word set corresponding to the risk sample document, and using the words in the risk word set as candidate keywords; a word frequency calculation unit, for calculating the word frequency of each candidate keyword in the risk sample document; a weight calculation unit, for calculating the overall position weight and relative position weight of each candidate keyword in the risk sample document; wherein the overall position weight is determined according to the ratio of the number of words in the risk sample document to the first word distance of the candidate keyword, the first word distance is represented by the number of words between the candidate keyword and the first word of the risk sample document; the relative position weight is determined according to the minimum word distance between the candidate keyword and the word representing time in the risk sample document; an inverse frequency calculation unit, for calculating the inverse document frequency of the candidate keyword in combination with the overall position weight and the relative position weight; a screening unit, for using the product of the word frequency and the inverse document frequency as a reference value, screening out candidate keywords whose reference values ​​meet preset requirements from the risk word set as risk keywords, so as to construct a risk subject dictionary of the specified risk subject based on the risk keywords.

8. A risk identification device, characterized in that: The device comprises at least one processor and a memory for storing processor-executable instructions, wherein the instructions implement the steps of the method according to any one of claims 1 to 6 when executed by the processor.

Images