A candidate item priority determination method for a civil aviation system

Abstract

The application provides a candidate item priority determination method for a civil aviation system, and relates to the technical field of data processing. The method first acquires a basic push score matrix according to a historical data set, then modifies the basic push score matrix in combination with a score adjustment weight matrix corresponding to a target user to realize dynamic weight adjustment and acquire a target push score matrix corresponding to the target user. If first target user data corresponding to an influencing factor satisfies a matching condition corresponding to the influencing factor, the influencing factor is taken as a correlation factor corresponding to the target user to acquire a correlation factor set corresponding to the target user. A sum of target push scores of all correlation factors in the correlation factor set corresponding to a candidate item is taken as a priority score corresponding to the candidate item. While retaining the global statistical advantage, the method significantly improves the personalization degree and the discrimination degree of the push, and enhances the discrimination degree and the accuracy of the priority of the candidate item corresponding to different users.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for determining the priority of candidates for civil aviation systems. Background Technology

[0002] In existing civil aviation systems, the recommendation order of candidate information is usually sorted according to the priority of the candidate options, with higher priority candidate information being displayed first to improve user interaction efficiency and recommendation effectiveness. In existing technologies, the priority of candidate options can be determined in the following ways: First, a fixed priority is obtained by statistically analyzing the user behavior data of all users within a historical time period; second, a personalized priority is obtained by predicting the strength of preference for candidate options based on the personalized historical behavior data of the target user through collaborative filtering or deep learning models.

[0003] However, the above method also has the following technical problems:

[0004] Candidate priority obtained through statistical analysis of user behavior data from all users over a historical period is global and can reflect overall trends, but it ignores individual differences between users and makes it difficult to achieve personalized recommendations. While predicting the preference intensity of target users for each candidate based on their personalized historical behavior data through collaborative filtering or deep learning models can generate more personalized recommendations, the personalized behavior data of target users is often sparse, resulting in small differences in priority between different users, which affects the discriminativeness and accuracy of the recommendation results. Summary of the Invention

[0005] To address the aforementioned technical problems, the technical solution adopted by this invention is as follows:

[0006] A method for determining the priority of candidates in a civil aviation system, comprising the following steps:

[0007] S10. Obtain the basic push score matrix A based on the historical dataset L; the element a in the i-th row and j-th column of A. ij Let X be the i-th influencing factor. i For the j-th candidate H j The basic push score; 1≤i≤m, where m is the number of influencing factors; 1≤j≤n, where n is the number of candidate options; L includes all data related to each candidate option collected within the historical time period.

[0008] S20. For each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching condition corresponding to the influencing factor, then the influencing factor is used as the associated factor corresponding to the target user Y to obtain the associated factor set F corresponding to Y; the first target user data is obtained based on the data acquisition rules corresponding to the matching condition corresponding to the influencing factor.

[0009] S30. Based on the second target user dataset corresponding to the preset behavior item, the judgment condition corresponding to the preset behavior item, and the user label corresponding to Y, obtain the score adjustment weight matrix C corresponding to Y; the element c in the i-th row and j-th column of C... ij For a ij The corresponding scores are adjusted with weights; the second target user dataset includes n second target user data points that correspond one-to-one with the candidate items; the second target user data is obtained based on the data acquisition rules corresponding to the preset behavior items.

[0010] S40. Based on A and C, obtain the target push score matrix D corresponding to Y; the element d in the i-th row and j-th column of D. ij For X i For H j Target push score; d ij =a ij ×c ij .

[0011] S50. For each candidate, the sum of the target push scores of all related factors in F is used as the priority score of the candidate.

[0012] The present invention has at least the following beneficial effects:

[0013] This invention provides a method for determining the priority of candidate options in a civil aviation system. The method obtains a basic push score matrix based on historical datasets. For each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching condition corresponding to the influencing factor, the influencing factor is used as a related factor corresponding to the target user to obtain a set of related factors corresponding to the target user. Based on the second target user dataset corresponding to a preset behavior item, the judgment condition corresponding to the preset behavior item, and the user tag corresponding to the target user, a score adjustment weight matrix corresponding to the target user is obtained. Based on the basic push score matrix and the score adjustment weight matrix corresponding to the target user, a target push score matrix corresponding to the target user is obtained, where the target push score matrix includes the target push score of each influencing factor for each candidate option. For each candidate option, the sum of the target push scores of all related factors in the set of related factors for the candidate option is used as the priority score corresponding to the candidate option. As can be seen, this invention first obtains a basic push score matrix based on historical datasets, and then modifies the basic push score matrix by combining it with the score adjustment weight matrix corresponding to the target user, thereby achieving dynamic weight adjustment to obtain the target push score matrix corresponding to the target user; it uses a dynamic weight matrix and a multi-source data fusion mechanism to obtain the target push score matrix; for each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching condition corresponding to the influencing factor, then the influencing factor is used as the associated factor corresponding to the target user to obtain the associated factor set corresponding to the target user; then the sum of the target push scores of all associated factors in the associated factor set is used as the priority score corresponding to the candidate option; while retaining the global statistical advantages, it significantly improves the personalization and distinguishability of the push, enhances the distinguishability and accuracy of the priority of the candidate options corresponding to different users, and further helps to improve the distinguishability and accuracy of the recommendation results. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a method for determining the priority of candidates in a civil aviation system, provided as an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar tasks and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] Embodiments of the present invention provide a method for determining the priority of candidates in a civil aviation system, the method comprising the following steps, such as... Figure 1 As shown:

[0019] S10. Obtain the basic push score matrix A based on the historical dataset L; the element a in the i-th row and j-th column of A. ij Let X be the i-th influencing factor. i For the j-th candidate H j The basic push score; 1≤i≤m, where m is the number of influencing factors; 1≤j≤n, where n is the number of candidate options; L includes all data related to each candidate option collected within the historical time period; such as: clicks, views, favorites, purchases, ratings, searches, time of occurrence of the behavior, geographical location, device type, network environment, age, gender, occupation, likes, favorites, etc.

[0020] Specifically, the end time of the historical time period is earlier than the current time, and the duration of the historical time period is the preset first duration.

[0021] Specifically, the candidates are items that may be recommended or displayed to users, such as products, articles, videos, and services.

[0022] Specifically, influencing factors are those that affect the order in which relevant information about the candidates is pushed, such as: hotel preferences, car rental preferences, severe weather, browsing preferences, flight preferences, evening peak hours, and flight itineraries including the top 10 cities for car rentals.

[0023] Specifically, -10≤a ij ≤10.

[0024] Furthermore, a ij The larger X is i For H j The greater the impact.

[0025] Specifically, step S10 includes the following steps S11-S12:

[0026] S11. Input the historical dataset into the g-th preset recommendation model to obtain X. i For H j The g-th initial push score W g ij 1≤g≤h, where h is the number of preset recommendation models.

[0027] Specifically, -10≤W g ij ≤10.

[0028] Specifically, the preset recommendation model is a recommendation model pre-determined from existing recommendation models. Different preset recommendation models use different recommendation algorithms, such as: content-based preset recommendation models, collaborative filtering-based recommendation models, and rule-based recommendation models.

[0029] S12, based on W g ij Get a ij a ij Meets the following conditions:

[0030] a ij =∑ h g=1 (V g ×W g ij V g is the preset importance weight corresponding to the g-th preset recommendation model.

[0031] Specifically, ∑ h g=1 V g =1.

[0032] Through the above steps, an initial push score is obtained based on historical datasets and a preset recommendation model. A basic push score is then obtained based on the preset importance weights corresponding to the recommendation model and the initial push score. The fusion of multiple models avoids the bias of a single model and helps to improve the accuracy of the obtained basic push score.

[0033] In one specific embodiment, those skilled in the art can obtain the basic push score of each influencing factor for each candidate in other ways, such as: experts setting the basic push score of each influencing factor for each candidate based on experience, prior knowledge of the system, or expert knowledge.

[0034] S20. For each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching condition corresponding to the influencing factor, then the influencing factor is used as the associated factor corresponding to the target user Y to obtain the associated factor set F corresponding to Y; the first target user data is obtained based on the data acquisition rules corresponding to the matching condition corresponding to the influencing factor.

[0035] Specifically, the primary target user data is obtained based on the data acquisition rules corresponding to the matching conditions of the influencing factors, including:

[0036] For each influencing factor, according to the data acquisition rules corresponding to the matching conditions of the influencing factor, the first target user data corresponding to the influencing factor is obtained from all data related to Y.

[0037] Specifically, the influencing factors, the matching conditions corresponding to the influencing factors, and the data acquisition rules corresponding to the matching conditions are stored in the first data configuration table corresponding to the influencing factors. The first data configuration table is a configuration table pre-set by those skilled in the art according to actual needs, including each influencing factor, the matching conditions corresponding to each influencing factor, and the data acquisition rules corresponding to each matching condition. For example, if the influencing factor is severe conditions, the matching condition corresponding to the influencing factor is: the current weather type is any one of heavy snow, freezing, low temperature, strong wind, blowing sand, and heavy rainfall; the data acquisition rule corresponding to the matching condition is: obtain the weather type of the current destination at the arrival time of the trip.

[0038] In this embodiment, not all influencing factors will affect the priority of the candidate options. For example, if the influencing factor is severe weather, but there is no severe weather at the departure point, arrival point, or transit point of the target user's current trip, then the influencing factor will not affect the priority of the candidate options. Through the above steps, the influencing factor is only used as a correlation factor when the first target user data meets the matching conditions corresponding to the influencing factor, thus avoiding interference from irrelevant factors.

[0039] S30. Based on the second target user dataset corresponding to the preset behavior item, the judgment condition corresponding to the preset behavior item, and the user label corresponding to Y, obtain the score adjustment weight matrix C corresponding to Y; the element c in the i-th row and j-th column of C... ij For a ijThe corresponding scores are adjusted with weights; the second target user dataset includes n second target user data points that correspond one-to-one with the candidate items; the second target user data is obtained based on the data acquisition rules corresponding to the preset behavior items.

[0040] Specifically, 0≤c ij ≤10.

[0041] Specifically, the preset behavior items and the corresponding judgment conditions are behavior items and judgment conditions preset by those skilled in the art according to actual needs. For example, the preset behavior item is exposure, and the judgment condition corresponding to the preset behavior item is: no click after ten exposures, no click after five exposures; or the preset behavior items are a first exposure-to-click ratio and a second exposure-to-click ratio, and the judgment condition corresponding to the first exposure-to-click ratio is: the ratio of the number of exposures to the number of clicks is between 0.3 and 0.5, and the judgment condition corresponding to the second exposure-to-click ratio is: the ratio of the number of exposures to the number of clicks is between 0.6 and 0.8.

[0042] Furthermore, the number of preset behavior items is not less than 1.

[0043] Specifically, the number of judgment conditions corresponding to the preset behavior item is not less than 1.

[0044] Specifically, the number of user tags corresponding to Y is no less than 1.

[0045] Specifically, the second target user data is obtained based on the data acquisition rules corresponding to the preset behavior items, and includes the following steps S01-S02:

[0046] S01. For each preset behavior item, obtain the data acquisition time period corresponding to the preset behavior item. The end time of the data acquisition time period is the current time point, and the duration of the data acquisition time period is the preset duration corresponding to the preset behavior item.

[0047] Preferably, the end time of the historical time period is earlier than the start time of the data acquisition time period corresponding to any preset behavior item.

[0048] S02. According to the data acquisition rules corresponding to the preset behavior item, collect data related to Y and H from the data acquisition time period corresponding to the preset behavior item. j From all data that have a correlation relationship, obtain the second target user dataset corresponding to the preset behavior item and H. j The corresponding second target user data.

[0049] Specifically, the data acquisition rules corresponding to the preset behavior items are stored in the second data configuration table corresponding to the preset behavior items. The second data configuration table is a configuration table pre-set by those skilled in the art according to actual needs, including each preset behavior item and the data acquisition rules corresponding to each preset behavior item.

[0050] By following the steps above, the data acquisition time period corresponding to each preset behavior item is obtained. Based on the data acquisition time period and the data acquisition rules corresponding to the preset behavior item, the second target user data corresponding to the preset behavior item is obtained. This can filter out irrelevant or invalid data, avoiding the waste of resources by processing irrelevant or invalid data.

[0051] Specifically, step S30 includes the following sub-steps S31-S33:

[0052] S31. For each preset behavior item, if the second target user dataset corresponding to the preset behavior item contains data that matches H... j If the corresponding second target user data meets the judgment condition corresponding to the preset behavior item, then according to the weight update rule corresponding to the judgment condition, the H value of each influencing factor corresponding to the preset behavior item is updated. j The basic push score is updated with the corresponding basic adjustment weight to update E; E is the basic adjustment matrix, and the element in the i-th row and j-th column of E is a. ij The corresponding basic adjustment weights.

[0053] Specifically, the weight update rules corresponding to the judgment conditions are update rules pre-set by those skilled in the art according to actual needs, such as: multiply by 0.2; multiply by 6; multiply by 3; multiply by 0.5; add 2; add 5, subtract 2; and so on.

[0054] Specifically, each update of the basic adjustment weight is based on the latest basic adjustment weight obtained from the previous update. According to the priority or execution order of the preset behavior items, the basic adjustment weight corresponding to the basic push score of the candidate item is iteratively adjusted for each influencing factor corresponding to the preset behavior item in turn, so as to ensure that the update process has time sequence and continuity. The priority or execution order of the preset behavior items is determined in advance by those skilled in the art according to actual needs, and will not be elaborated here.

[0055] Specifically, the initial value of the base adjustment weight is 1.

[0056] Specifically, the correspondence between preset behavioral items and influencing factors is determined in advance by those skilled in the art based on actual needs, and will not be elaborated here.

[0057] S32. Use the updated E as the first adjustment matrix G corresponding to M; the element in the i-th row and j-th column of G is a. ij The corresponding first adjustment weight.

[0058] S33. Update G according to the user label corresponding to Y to obtain C.

[0059] Through the above steps, the basic adjustment matrix is ​​updated and iterated multiple times based on the preset behavior items and the second target user data corresponding to the preset behavior items to obtain the first adjustment weight matrix. The first adjustment weight matrix is ​​then updated based on the user tags corresponding to the target users to obtain the score adjustment weight matrix corresponding to the target users. Determining the score adjustment weight matrix based on the characteristics of the target users helps to improve the personalization and distinguishability of the obtained score adjustment weight matrix.

[0060] Specifically, step S33 includes the following steps S331-S332:

[0061] S331. For each user tag corresponding to Y, if the user tag is the same as any preset tag, then according to the weight update rule corresponding to the preset tag that is the same as the user tag, the first adjustment weight corresponding to the basic push score of each candidate for each influencing factor corresponding to the user tag is updated so as to update G.

[0062] Specifically, the weight update rules corresponding to the preset labels are update rules pre-set by those skilled in the art according to actual needs, such as: multiply by 0.2; multiply by 6; multiply by 3; multiply by 0.5; add 2; add 5, subtract 3; set to 4; set to 0; and so on.

[0063] Specifically, each update of the first adjustment weight is based on the latest first adjustment weight obtained from the previous update. According to the priority or execution order of user tags, the first adjustment weight corresponding to the basic push score of each influencing factor of the candidate is iteratively adjusted in turn to ensure that the update process is sequential and continuous. The priority or execution order of user tags is determined in advance by those skilled in the art according to actual needs, and will not be elaborated here.

[0064] Specifically, the influencing factors corresponding to the user tag are the same as those corresponding to the preset tags that are the same as the user tag.

[0065] Specifically, the correspondence between preset labels and influencing factors is pre-set by those skilled in the art according to actual needs, and will not be elaborated here.

[0066] S332, G after all updates are completed is C.

[0067] By performing the above steps, the first adjusted weight matrix is ​​updated and iterated multiple times based on the user tags of the target user to obtain the score adjusted weight matrix corresponding to the target user. This helps to improve the personalization and distinguishability of the obtained score adjusted weight matrix.

[0068] S40. Based on A and C, obtain the target push score matrix D corresponding to Y; the element d in the i-th row and j-th column of D. ij For X i For H j Target push score; d ij =a ij ×c ij .

[0069] S50. For each candidate, the sum of the target push scores of all related factors in F is used as the priority score corresponding to the candidate.

[0070] Specifically, the higher the priority score of a candidate, the higher the priority of that candidate.

[0071] Furthermore, the candidate priority is used to determine the order in which relevant information about the candidate is pushed, wherein information about the candidate with higher priority is pushed to M first.

[0072] Through the above steps, firstly, a basic push score matrix is ​​obtained based on historical datasets; for each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching conditions corresponding to the influencing factor, then the influencing factor is used as a related factor corresponding to the target user to obtain a set of related factors corresponding to the target user; the influence of irrelevant influencing factors is avoided; based on the second target user dataset corresponding to the preset behavior item, the judgment conditions corresponding to the preset behavior item, and the user tags corresponding to the target user, a score adjustment weight matrix corresponding to the target user is obtained; based on the basic push score matrix and the score adjustment weight matrix corresponding to the target user, a target push score matrix corresponding to the target user is obtained. The target push score matrix includes the target push score for each influencing factor for each candidate option. The basic push score matrix is ​​modified by adjusting the weight matrix based on the score corresponding to the target user. A dynamic weight matrix and multi-source data fusion mechanism are used to obtain the target push score matrix for the target user. The sum of the target push scores of all related factors in the related factor set for the candidate option is used as the priority score for that candidate option. While retaining the advantages of global statistics, this significantly improves the personalization and distinguishability of the push, enhances the distinguishability and accuracy of the priority of candidate options for different users, and further improves the distinguishability and accuracy of the recommendation results.

[0073] In one specific embodiment, the following steps are included after step S32:

[0074] S321. Based on the data validity judgment conditions corresponding to the preset behavior item, determine whether each second target user data corresponding to the preset behavior item is valid data.

[0075] Specifically, the data validity judgment conditions corresponding to the preset behavior items are conditions pre-set by those skilled in the art based on actual needs to determine whether the data is valid. For example, if the preset behavior item is exposure, the data validity judgment corresponding to the preset behavior item is that the exposure duration is greater than the preset minimum exposure duration. If the data in the second target user data corresponding to the preset behavior item indicates that the exposure duration is not greater than the preset minimum exposure duration, then the second target user data is determined to be invalid data. If the data in the second target user data corresponding to the preset behavior item indicates that the exposure duration is greater than the preset minimum exposure duration, then the second target user data is determined to be valid data.

[0076] S322, For H j If in the second target user dataset corresponding to all preset behavior items, there is a match with H j If all the corresponding second target user data are invalid, then H is determined. j Data label H_G in G j Invalid; otherwise, determine H_G j It is valid.

[0077] S323, According to G, H_G j Get C from the user tag corresponding to Y.

[0078] In this embodiment, when the target user is a new user, or when the target user unintentionally clicks or scrolls over certain content, some invalid data may be generated. Through the above steps, the validity of the second target user data is determined according to the data validity judgment conditions corresponding to the preset behavior items. If the second target user data corresponding to the candidate items in all the second target user data sets corresponding to the preset behavior items are invalid data, then the first adjustment weights corresponding to the basic push scores of the candidate items obtained from these invalid second target user data are also invalid. Therefore, the data labels of the candidate items in the first adjustment weight matrix are determined to be invalid. Otherwise, the data labels of the candidate items in the first adjustment weight matrix are determined to be valid. Obtaining the score adjustment weight matrix corresponding to the target user based on the data labels of the candidate items in the first adjustment weight matrix helps to improve the accuracy of the score adjustment weight matrix.

[0079] Specifically, S323 also includes the following steps:

[0080] S100, when at least α candidate options have valid data labels in G, and not all candidate options have valid data labels in G, if H_G j If invalid, then determine H. j It is the first candidate; if H_G j For it to be valid, then determine H. j The second candidate; to obtain the list of first candidate options L=(L1, L2, ..., L... k , ..., L t The second candidate list P = (P1, P2, ..., P...) y , ..., P q ); α is the preset number of valid items; L k Let P be the k-th first candidate, 1≤k≤t, where t is the number of first candidates. y Let y be the y-th second candidate, 1≤y≤q, where q is the number of second candidates.

[0081] Specifically, 1 < α < n.

[0082] Specifically, t+q=n.

[0083] Specifically, step S100 further includes: if the data labels of each candidate in G are all valid or all invalid, then G is used as R.

[0084] S200. From the first adjustment matrix corresponding to all other users obtained within the preset time period, select L. k The corresponding first intermediate matrix list Q k =(Q k1 Q k2 Q kr , ...Q ks(k) ), Q kr For L k The corresponding r-th first intermediate matrix, 1≤r≤s(k), where s(k) is L k The number of corresponding first intermediate matrices; P1, P2, ..., P y , ..., P q and L k In Q kr All data labels in Q are valid; this can be understood as: Q kr It belongs to the set constructed from the first adjustment matrix corresponding to all other users obtained from the preset time period.

[0085] Specifically, "other users" refers to any user other than the target user.

[0086] Specifically, the end time of the preset time period is the current time, the duration of the preset time period is the preset second duration, and the preset second duration is less than the preset first duration; preferably, the end time of the preset time period is the current time.

[0087] S300, If the second intermediate matrix G corresponds to G... 0 With Q kr The corresponding third intermediate matrix Q 0 kr The similarity between them is greater than the preset similarity threshold S 0 Then Q kr As L k The corresponding first key matrix is ​​used to obtain L k The corresponding first key matrix list V k =(V k1 V k2 , ..., V kz , ..., V ku(k) V kz For L k The corresponding z-th first key matrix, 1≤z≤u(k), where u(k) is L k The number of corresponding first key matrices; G 0 and Q 0 kr Both are m×q matrices and G 0 The element in row i and column y of G is X. i For P y The first adjustment weight corresponding to the basic push score, Q 0 kr The element in the i-th row and y-th column is Q. kr ZhongX i For P y The first adjustment weight corresponding to the basic push score; those skilled in the art know that any method for obtaining the similarity between two matrices in the prior art is within the protection scope of this invention, and will not be described in detail here.

[0088] Specifically, G 0 With Q 0 kr The greater the similarity between them, the better G 0 With Q 0 kr The more similar they are.

[0089] Specifically, 0.8≤S 0 ≤1.

[0090] S400, V k1 V k2 , ..., V kz , ..., V ku(k) ZhongXi For L k The weighted average of the first adjusted weights corresponding to the basic push score is used as X in G. i For L k The first adjustment weight corresponding to the basic push score is used to update G and then use the updated G as the second adjustment matrix R corresponding to M.

[0091] S500, Update R according to the user tag corresponding to Y to obtain C.

[0092] Through the above steps, when at least α candidate options have valid data labels in the first adjusted weight matrix corresponding to the target user, and not all candidate options have valid data labels in the first adjusted weight matrix, the first adjusted weight of the target user can be updated according to the first adjusted weights corresponding to other users. Candidates with invalid data labels in the first adjusted weight matrix corresponding to the target user are designated as first candidates, and candidates with valid data labels in the second adjusted weight matrix corresponding to the target user are designated as second candidates. For each first candidate, a first intermediate layer is selected from the first adjusted weight matrix corresponding to other users, where the data labels of the first candidate and all second candidates are valid. The matrix, based on the similarity between the second and third intermediate matrices, selects the data labels of the first candidate as valid and relatively similar to the first adjustment matrix corresponding to the target user as the first key matrix. Based on the first adjustment weight corresponding to the basic push score of the candidate for each influencing factor in the first key matrix, the first adjustment weight corresponding to the basic push score of the candidate for each influencing factor in the first adjustment weight matrix corresponding to the target user is iteratively updated; and the first adjustment weight matrix after all updates is used as the second adjustment weight matrix corresponding to the target user; a compensation mechanism is adopted to improve the accuracy of the second adjustment weight matrix, and further improves the accuracy of the score adjustment weight matrix.

[0093] Specifically, step S500 also includes the following steps:

[0094] S501. For each user tag corresponding to Y, if the user tag is the same as any preset tag, then according to the weight update rule corresponding to the preset tag that is the same as the user tag, the second adjustment weight corresponding to the basic push score of each candidate for each influencing factor corresponding to the user tag is updated so as to update R.

[0095] Specifically, each update of the second adjustment weight is based on the latest second adjustment weight obtained from the previous update. According to the priority of user tags or the execution order, the second adjustment weight corresponding to the basic push score of each influencing factor of the user tag is iteratively adjusted to ensure that the update process is sequential and continuous.

[0096] S502, Use the updated R as C.

[0097] By performing the above steps, the second adjusted weight matrix is ​​updated and iterated multiple times based on the user tags of the target user to obtain the score adjusted weight matrix corresponding to the target user. This helps to improve the personalization and distinguishability of the obtained score adjusted weight matrix.

[0098] In one specific embodiment, after step S200 and before step S500, the following step is further included to obtain R:

[0099] S210. If the first specified matrix G corresponds to G... 1 With Q kr The corresponding second specified matrix Q 1 kr The similarity between them is greater than S 0 Then Q kr As L k The corresponding second key matrix is ​​used to obtain L k The corresponding second key matrix list W k =(W k1 W k2 ,…,W kβ ,…,W kθ(k) ), W kβ For L k The corresponding β-th second key matrix, 1≤β≤θ(k), where θ(k) is L k The number of corresponding second key matrices; G 1 and Q 1 kr Both are f×q matrices and G 1 The element in row b and column y of G is F. b For P y The first adjustment weight corresponding to the basic push score, Q 1 kr The element in the b-th row and y-th column is Q. kr China F b For P y The first adjustment weight corresponding to the basic push score, where f is the number of related factors in F, F b Let f be the b-th related factor in F, where 1 ≤ b ≤ f.

[0100] Specifically, G1 With Q 1 kr The greater the similarity between them, the better G 1 With Q 1 kr The more similar they are.

[0101] S220, W k1 W k2 ,…,W kβ ,…,W kθ(k) China F b For L k The weighted average of the first adjusted weights corresponding to the basic push score is used as F in G. b For L k The first adjustment weight corresponding to the basic push score is used to update G and take the updated G as R.

[0102] Through the above steps, since not all influencing factors affect the priority of candidate options, based on the similarity between the first and second specified matrices, the data labels of the first candidate option can be selected as valid second key matrices that are similar to the first adjustment matrix corresponding to the target user. Based on the first adjustment weight corresponding to the basic push score of the candidate option for each related factor in the second key matrix, the first adjustment weight corresponding to the basic push score of the candidate option for each related factor in the first adjustment weight matrix corresponding to the target user is iteratively updated; and the first adjustment weight matrix after all updates is used as the second adjustment weight matrix corresponding to the target user. Compared with the above embodiment, the amount of data that needs to be processed is reduced, which is conducive to improving computational efficiency. At the same time, it can also reduce the influence of irrelevant influencing factors, which is conducive to improving the accuracy of the second adjustment weight matrix, and further, it is conducive to improving the accuracy of the score adjustment weight matrix.

[0103] In one specific embodiment, the following steps are included after step S50:

[0104] S60. For each of the several candidate adjustment rules, obtain the dataset corresponding to the candidate adjustment rule. The dataset includes the question text, at least one influencing factor, at least one candidate option, and the score adjustment method.

[0105] Specifically, the candidate adjustment rules are predetermined by those skilled in the art based on actual needs, such as: recommendation rules for enhanced car service.

[0106] Specifically, the question text includes: a query text; the query text is a text pre-set by a person skilled in the art to clarify the decision-making objective according to the candidate adjustment rule, for example: the candidate adjustment rule is a car service enhancement recommendation rule, and the query text corresponding to the candidate adjustment rule includes the question: whether to adjust the priority score of the candidate options related to car service.

[0107] Specifically, the question text also includes a prompt text, which is used to guide the large language model to obtain information related to the decision-making objective corresponding to the question text. This helps the large language model understand which external data should be collected for auxiliary judgment. For example, based on factors such as the current user's geographical location, local car use policies, and whether there are major events or holidays, please comprehensively judge whether it is necessary to increase the priority score of the candidate options related to car use services.

[0108] Specifically, after receiving the question text, the large language model automatically calls external data sources, such as weather APIs, map APIs, and local databases, to integrate and reason about the information and output the decision result corresponding to the question text.

[0109] Furthermore, the correspondence between the candidate adjustment rules and the influencing factors, as well as the correspondence between the candidate adjustment rules and the candidate options, are predetermined by those skilled in the art based on actual needs. For example, the candidate adjustment rule is a car service enhancement recommendation rule, and the influencing factors corresponding to the candidate adjustment rule are flight itineraries including the top 10 cities for car services and evening peak hours; the candidate options corresponding to the candidate adjustment rule are vehicle reservation services, real-time traffic information push services, and carpooling services.

[0110] Furthermore, the score adjustment method corresponding to the candidate adjustment rule is an adjustment method pre-set by those skilled in the art based on the candidate adjustment rule. For example, if the candidate adjustment rule is a car service enhancement recommendation rule, the score adjustment method corresponding to the candidate adjustment rule is: adjust the priority score to 100; adjust the priority score to 90; and so on.

[0111] S70. For each candidate adjustment rule, if F includes all influencing factors corresponding to the candidate adjustment rule, then input the question text corresponding to the candidate adjustment rule into the large language model to obtain the decision result J corresponding to the question text.

[0112] S80. If J is a first-class result, then the priority scores of all candidates corresponding to the candidate adjustment rule are adjusted according to the score adjustment method corresponding to the candidate adjustment rule corresponding to J, so as to obtain the target priority score corresponding to each candidate; the first-class result indicates that the priority of the candidate corresponding to the candidate adjustment rule needs to be adjusted.

[0113] Specifically, the decision results also include a second type of result, which indicates that there is no need to adjust the priority of the candidates corresponding to the candidate adjustment rules.

[0114] Specifically, the higher the target priority score of a candidate, the higher the priority of that candidate.

[0115] Through the above steps, compared to manually adjusting the priority scores of candidate options based on actual needs, the embodiments of this application achieve priority score adjustment through a large language model. In the embodiments of this application, a dataset corresponding to the candidate adjustment rule is obtained. The dataset includes the question text, at least one influencing factor, at least one candidate option, and a score adjustment method. If the set of related factors includes all influencing factors corresponding to the candidate adjustment rule, the question text corresponding to the candidate adjustment rule is input into the large language model to obtain the decision result corresponding to the question text. If the decision result is a first type of result, the priority scores of all candidate options corresponding to the candidate adjustment rule are adjusted according to the score adjustment method corresponding to the candidate adjustment rule corresponding to the decision result to obtain the target priority score for each candidate option. This achieves automated and intelligent adjustment of the candidate option priority scores, which not only improves the accuracy of the target priority score but also effectively reduces manual maintenance costs and significantly improves data processing efficiency.

[0116] Embodiments of the present invention also provide a non-transitory computer-readable storage medium that can be disposed in an electronic device to store a computer program related to implementing a method in the method embodiments, the computer program being loaded and executed by the processor to implement the method provided in the above embodiments.

[0117] Embodiments of the present invention also provide an electronic device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method provided in the above embodiments.

[0118] Embodiments of the present invention also provide a computer program product including program code, which, when the program product is run on an electronic device, causes the electronic device to perform the steps of the methods described above in various exemplary embodiments of the present invention.

[0119] This invention provides a method for determining the priority of candidate options in a civil aviation system. The method obtains a basic push score matrix based on historical datasets. For each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching condition corresponding to the influencing factor, the influencing factor is used as a related factor corresponding to the target user to obtain a set of related factors corresponding to the target user. Based on the second target user dataset corresponding to a preset behavior item, the judgment condition corresponding to the preset behavior item, and the user tag corresponding to the target user, a score adjustment weight matrix corresponding to the target user is obtained. Based on the basic push score matrix and the score adjustment weight matrix corresponding to the target user, a target push score matrix corresponding to the target user is obtained, where the target push score matrix includes the target push score of each influencing factor for each candidate option. For each candidate option, the sum of the target push scores of all related factors in the set of related factors for the candidate option is used as the priority score corresponding to the candidate option. As can be seen, this invention first obtains a basic push score matrix based on historical datasets, and then modifies the basic push score matrix by combining it with the score adjustment weight matrix corresponding to the target user, thereby achieving dynamic weight adjustment to obtain the target push score matrix corresponding to the target user; it uses a dynamic weight matrix and a multi-source data fusion mechanism to obtain the target push score matrix; for each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching condition corresponding to the influencing factor, then the influencing factor is used as the associated factor corresponding to the target user to obtain the associated factor set corresponding to the target user; then the sum of the target push scores of all associated factors in the associated factor set is used as the priority score corresponding to the candidate option; while retaining the global statistical advantages, it significantly improves the personalization and distinguishability of the push, enhances the distinguishability and accuracy of the priority of the candidate options corresponding to different users, and further helps to improve the distinguishability and accuracy of the recommendation results.

[0120] While specific embodiments of the invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and are not intended to limit the scope of the invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the invention.

Claims

1. A method for determining the priority of candidate options in a civil aviation system, characterized in that, The method includes the following steps: S10. Obtain the basic push score matrix A based on the historical dataset L; the element a in the i-th row and j-th column of A. ij Let X be the i-th influencing factor. i For the j-th candidate H j The basic push score; 1≤i≤m, where m is the number of influencing factors; 1≤j≤n, where n is the number of candidate options; L includes all data related to each candidate option collected within the historical time period; S20. For each influencing factor, if the first target user data corresponding to the influencing factor satisfies the matching condition corresponding to the influencing factor, then the influencing factor is used as the associated factor corresponding to the target user Y to obtain the associated factor set F corresponding to Y; the first target user data is obtained based on the data acquisition rules corresponding to the matching condition corresponding to the influencing factor. S30. Based on the second target user dataset corresponding to the preset behavior item, the judgment condition corresponding to the preset behavior item, and the user label corresponding to Y, obtain the score adjustment weight matrix C corresponding to Y; the element c in the i-th row and j-th column of C... ij For a ij The corresponding score will be adjusted accordingly. The second target user dataset includes n second target user data points that correspond one-to-one with the candidate items; the second target user data is obtained based on the data acquisition rules corresponding to the preset behavior items; step S30 includes the following sub-steps: S31. For each preset behavior item, if the second target user dataset corresponding to the preset behavior item contains data that matches H... j If the corresponding second target user data meets the judgment condition corresponding to the preset behavior item, then according to the weight update rule corresponding to the judgment condition, the H value of each influencing factor corresponding to the preset behavior item is updated. j The basic push score is updated with the corresponding basic adjustment weight to update E; E is the basic adjustment matrix, and the element in the i-th row and j-th column of E is a. ij The corresponding basic adjustment weights; each update of the basic adjustment weights is based on the latest basic adjustment weights obtained from the previous update. According to the priority or execution order of the preset behavior items, the basic adjustment weights corresponding to the basic push scores of the candidate items are iteratively adjusted for each influencing factor corresponding to the preset behavior item. S32. Use the updated E as the first adjustment matrix G corresponding to M; the element in the i-th row and j-th column of G is a. ij The corresponding first adjustment weight; S33. Update G according to the user label corresponding to Y to obtain C; S40. Based on A and C, obtain the target push score matrix D corresponding to Y; the element d in the i-th row and j-th column of D. ij For X i For H j Target push score; d ij =a ij ×c ij ; S50. For each candidate, the sum of the target push scores of all related factors in F is used as the priority score corresponding to the candidate.

2. The method for determining the priority of candidates for a civil aviation system according to claim 1, characterized in that, The primary target user data is obtained based on the data acquisition rules corresponding to the matching conditions of the influencing factors, specifically including: For each influencing factor, according to the data acquisition rules corresponding to the matching conditions of the influencing factor, the first target user data corresponding to the influencing factor is obtained from all data related to Y.

3. The method for determining the priority of candidates for a civil aviation system according to claim 1, characterized in that, The second target user data is obtained based on the data acquisition rules corresponding to preset behavior items, specifically including the following steps: S01. For each preset behavior item, obtain the data acquisition time period corresponding to the preset behavior item. The end time of the data acquisition time period is the current time point, and the duration of the data acquisition time period is the preset duration corresponding to the preset behavior item. S02. According to the data acquisition rules corresponding to the preset behavior item, collect data related to Y and H from the data acquisition time period corresponding to the preset behavior item. j From all data that have a correlation relationship, obtain the second target user dataset corresponding to the preset behavior item and H. j The corresponding second target user data.

4. The method for determining the priority of candidates for a civil aviation system according to claim 1, characterized in that, Step S33 includes the following steps: S331. For each user tag corresponding to Y, if the user tag is the same as any preset tag, then according to the weight update rule corresponding to the preset tag that is the same as the user tag, the first adjustment weight corresponding to the basic push score of each candidate for each influencing factor corresponding to the user tag is updated so as to update G. S332, G after all updates are completed is C.

5. The method for determining the priority of candidates in a civil aviation system according to claim 1, characterized in that, The higher the priority score of a candidate option, the higher its priority.

6. The method for determining the priority of candidates for a civil aviation system according to claim 5, characterized in that, Candidate priority is used to determine the order in which relevant information about the candidate is pushed. Among them, information about the candidate with higher priority is pushed to M first.

7. The method for determining the priority of candidates for a civil aviation system according to claim 1, characterized in that, -10≤a ij ≤10。 8. The method for determining the priority of candidates for a civil aviation system according to claim 1, characterized in that, 0≤c ij ≤10。 9. The method for determining the priority of candidates in a civil aviation system according to claim 1, characterized in that, The initial value of the base adjustment weight is 1.

Images