Flight meal personalized recommendation method
By combining passenger itinerary data and historical meal selection information, using technical means such as vectorization and Pearson correlation coefficients to generate personalized meal recommendation information and provide intelligent search functions, the problem of rigid food selection mechanism in the existing system is solved, and the satisfaction of passengers' personalized needs and improvement of service experience is achieved.
Patent Information
- Application Number
- CN202411804611.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-10
- Publication Date
- 2025-04-25
AI Technical Summary
The existing flight meal booking system is relatively rigid in the meal selection mechanism, lacking intuitive search functions and customized recommendation services, which makes it difficult for passengers to make choices based on their personal taste preferences when facing massive information, and cannot effectively meet personalized needs.
By combining passengers' all-round itinerary data and historical meal selection information, a package database and a passenger database are established, and technical means such as vectorization and Pearson correlation coefficient are used to generate personalized meal recommendation information, and provide intelligent search functions to tailor meal selection solutions.
It enables passengers to easily choose the package they want, reduces the redundancy of meal selection information, improves service experience, meets passengers' personalized needs, and provides intelligent search functions to accurately capture demand tendencies.
Smart Images
Figure CN119886630A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of personalized recommendation algorithms, and in particular relates to a personalized recommendation method for flight meals. Background Art
[0002] When a passenger's flight coincides with a time period when meals are required, the airline will provide passengers with corresponding meal services. However, in current airline practices, only some airlines provide passengers with meal selection services based on meal types and cabin classes, and this service is relatively basic and limited to listing the meal items available for selection.
[0003] The current flight meal reservation system relies on mini programs, official apps and other platforms to provide passengers with meal options based on a variety of package categories and cabin classes. However, the system is relatively rigid in its meal selection mechanism, lacking intuitive search functions and customized recommendation services. During the reservation process, passengers need to comprehensively review various meal information, including meal names, attractive pictures and detailed descriptions. Faced with massive amounts of information, it is often difficult for passengers to quickly make choices based on their personal taste preferences, which to a certain extent restricts the satisfaction of passengers' personalized needs and the improvement of service experience. In view of this, this solution innovatively integrates passengers' all-round itinerary data, including flight details, cabin class, user profile and past meal selection records, in order to create a personalized meal recommendation system. Summary of the invention
[0004] The purpose of the present invention is to provide a personalized recommendation method for flight meals, which combines data such as passenger itinerary information and historical meal selection information to actively push recommended meal selection plans, facilitate passengers to easily choose the desired meal packages, reduce the redundancy of passenger meal selection information, and provide passengers with intelligent search behavior, accurately capture their demand tendencies, and thus tailor meal selection plans.
[0005] To achieve the above object, the present invention provides a method for personalized recommendation of flight meals, comprising the following steps: S1. Database construction: Establishing a meal package database and a passenger database, wherein the meal package database includes a plurality of meal packages and meal package labels, wherein the meal package labels include meal prices, applicable cabins for meals, applicable membership levels for meals, whether meals are applicable to point redemption activities and point redemption prices, meal names, meal pictures, meal descriptions, meal contents, and meal labels; The passenger database includes passenger user information, passenger itinerary information, and historical meal selection information. The passenger user information includes historical itinerary information, membership level, and points balance; the passenger itinerary information includes the passenger's current flight information and cabin information; the historical itinerary information includes the passenger's historical flight information and historical cabin information; and the historical meal selection information includes historical meal package information selected corresponding to the historical itinerary information; S2. Generate fixed meal information: The fixed meal information is a meal list specified for passengers of different membership levels, points balances and flight levels. Based on the passenger user information and passenger itinerary information, combined with the meal tag of the meal database, the database query and combined output of the fixed meal information P are performed. ; S3, generating recommended meal information, including the following steps: S3.1. Obtain the passenger's historical meal information h, where both the historical meal information and the meal in the meal database contain multiple meal attributes; S3.2, vectorize the historical package information h and the package information in the package database S_meals, and the vectorization formula is:
[0006]
[0007] Among them, H is a vector matrix of K×u×N, S is a vector matrix of K×u×M, K represents the number of meal attributes, u represents the vector dimension size corresponding to a single meal attribute, N represents the number of historical packages, and M represents the number of packages in the package database; Represents a language model, which is used to vectorize package information; S3.3. The feature similarity C between the historical packages and the packages in the package database is obtained through the Pearson correlation coefficient. The calculation formula is: C PCorrelation
[0008] in, for vector matrix, K represents the number of meal attributes, M represents the number of meal packages in the meal package database, and N represents the number of historical meal packages. contains the feature similarity c between the meal attributes of each historical meal and each meal in the meal database, for The vector matrix of , Indicates shared Similarity matrix of historical meals.
[0009] S3.4. Obtain the similarity weight value corresponding to each meal attribute in the historical package, and calculate the The total recommendation score of each package in the package database based on the attributes of the historical packages , the calculation formula is:
[0010] in, Indicates The weight value corresponding to each meal attribute, K represents the number of meal attributes, Represents the M packages in the package database and the historical package Each meal attribute corresponds to a similarity score, which is an M-dimensional vector. The size is dimensional vector, , express Recommendation scores for historical meals; S3.5. Calculate the recommendation score F of each package in the package database based on the historical time. The calculation formula is:
[0011] in, is the weight of the corresponding time of the historical package. The closer the time of the historical package is to the current time, the greater the weight value. N represents the number of historical packages; S3.6. Sort the recommendation scores of all meal packages to obtain sorted meal information R. The sorting formula is:
[0012] in, The recommended scores of the packages are sorted from large to small, and the recommended scores of the packages involved in the sorting are all greater than 0.5, and then the mapping function is used Obtain sorted meal information R, where R is the recommended meal information; S4, comprehensive ranking: Combine the fixed meal information and recommended meal information, sort and adjust the package p in the package database, traverse all packages p, if p ,and , then sort according to the order of R's packages, and the sorted set is ;like ,and , then according to The order of packages is sorted, and the sorted set is ;like ,and , then sort according to the order of P's packages, and the sorted set is , the final order of meal push is ={ }, and then push the meal packages in the meal package database to the meal selection interface in the sorting order of T for passengers to choose; S5. Precise meal search: The meal selection interface has a search box. Passengers can enter search information I in the search box to obtain meal recommendation results. , the formula is as follows:
[0013] Among them, RNN is a sorting model, which obtains relevant packages according to the retrieval information I, and sorts the packages according to the sorting method of T and pushes them to passengers.
[0014] Further, in step S1, the meal content is the ingredients of the meal, the meal label is a description of the type, taste, and dishes of the meal; the flight information includes the flight number, flight date, departure place, destination, and travel time period.
[0015] Further, in step S3.1, the meal attributes include meal name, meal description, meal content, and meal label. The meal content is the composition of the ingredients of the meal, and each ingredient is an attribute.
[0016] Further, in step S3.6, F={ }, if the package with the highest recommendation score in F is , the second most recommended package is ,but .
[0017] Furthermore, in step 3.6, if the number of historical meal packages is 0, only the fixed meal information P is pushed to the passenger.
[0018] After adopting the above scheme, the beneficial effects of the present invention are: The present invention formulates fixed meal information based on passenger user information and passenger itinerary information, combined with the package label of the package database, and then obtains recommended meal information based on the passenger's historical meal selection information. The recommended meal information is calculated based on the meal attributes of each package in the historical package information and the feature similarity between each package in the package database, as well as the attribute weight and the time weight of the historical package. The feature similarity and time factor of the package are taken into account. The top-ranked package is most similar to the historical package with a time closer to the current time. Then, the sorting method of the fixed meal information and the recommended meal information is integrated to obtain the final sorting of the packages in the package database, and the packages are pushed to the passengers according to this sorting method. The top-ranked packages are more in line with the personal preferences of the passengers, which is convenient for the passengers to easily choose the desired packages and reduce the redundant information of the passengers' meal selection. In addition, on the basis of the final sorting of the recommended packages, the present invention also provides passengers with an intelligent search function, accurately captures their demand tendencies according to the search information input by the passengers, and thus tailors the package selection plan, so that the passengers experience convenience and personalized services in the meal ordering process. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a flowchart of the method of the present invention; Figure 2 A flowchart of generating recommended meal information for the present invention. DETAILED DESCRIPTION
[0020] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0021] The present invention provides a personalized recommendation method for flight meals, such as Figure 1 As shown, the specific steps include: S1. Database construction: A package database and a passenger database are established. The package database contains multiple packages. Passengers need to choose and book from these packages. The package database also contains the labels of the packages. The labels of the packages include meal prices, applicable cabins for meals, applicable membership levels for meals, whether meals are applicable to points redemption activities and points redemption prices, meal names, meal pictures, meal descriptions, meal contents, and meal labels. Among them, the meal contents are generally the ingredients of the package. For example, the ingredients of the beef fried rice package include beef, rice, eggs, peppers, apples, and yogurt. The meal label is a description of the type, taste, and dishes of the package, such as whether the package is Chinese or Western, whether the taste is sweet, sour, bitter, or spicy, whether it is a cold dish or a hot dish, etc. Taking the beef fried rice package as an example, the beef fried rice package is a Chinese package and a spicy hot dish.
[0022] The passenger database includes passenger user information, passenger itinerary information, and historical meal selection information. The passenger user information includes historical itinerary information, membership level, and points balance; the passenger itinerary information includes the passenger's current flight information and cabin information. The flight information includes flight number, flight date, departure place, destination, travel time period, etc., and the cabin information includes first class, business class, and economy class, etc.; the historical itinerary information includes the passenger's historical flight information and historical cabin information; the historical meal selection information includes historical package information selected corresponding to the historical itinerary information, and the historical package information includes the passenger's historical meal selections and the time of meal selections.
[0023] S2. Generate fixed meal information: The fixed meal information is a meal list specified by the airline for passengers of different membership levels, points balances and flight classes. For example, a fruit meal is provided free of charge to passengers of all membership levels, all travel time periods, and with a minimum points of 1,000 points and business class, or it can only be purchased by these passengers. Based on the passenger user information and passenger travel information, combined with the meal tag of the meal database, the database query and combined output of the fixed meal information P are performed. , thereby fixing the meal information as part of the meal set in the meal set database.
[0024] S3, generate recommended meal information, such as Figure 2 As shown, the specific steps include: S3.1. Obtain the historical meal information h of the passenger. The historical meal information and the meal in the meal database both contain multiple meal attributes. The meal attributes include meal name, meal description, meal content, and meal label. The meal content is the ingredients of the meal, and each ingredient is an attribute. S3.2, vectorize the historical package information h and the package information in the package database S_meals to obtain the corresponding vectors H and S, that is, digitize the text information of the package to facilitate the subsequent similarity calculation. The vectorization formula is:
[0025]
[0026] Among them, H is a vector matrix of K×u×N, S is a vector matrix of K×u×M, K represents the number of meal attributes, the number of meal attributes of the historical meal packages and the meal package database is the same, u represents the vector dimension size corresponding to a single meal attribute, N represents the number of historical meal packages, and M represents the number of meal packages in the meal package database, that is, H and S are both three-dimensional matrices of meal types, meal attributes of meal packages, and vector dimensions corresponding to meal attributes; Represents a language model, which is used to vectorize package information.
[0027] S3.3. The feature similarity C between the historical packages and the packages in the package database is obtained through the Pearson correlation coefficient. The calculation formula is: C PCorrelation
[0028] in, for vector matrix, K represents the number of meal attributes, M represents the number of meal packages in the meal package database, and N represents the number of historical meal packages. Contains the similarity between the meal attributes of each historical meal and the features of each meal in the meal database , for The matrix is the similarity matrix of all meal attributes of a historical meal set and each meal set in the meal database. , Indicates shared Similarity matrix of historical meals.
[0029] S3.4. Obtain the similarity weight value corresponding to each meal attribute in the historical set meal. For example, the ingredient attributes of the ham fried rice set meal include lunch beef, rice, corn and carrot. The most important and representative component of the ham fried rice set meal is lunch beef, so the corresponding weight value of lunch beef is the highest, followed by rice, and the weights of other attributes are lower; then calculate the The total recommendation score based on the attributes of the historical packages and each package in the package database , the calculation formula is:
[0030] in, Indicates The weight value corresponding to each meal attribute, K represents the number of meal attributes; Represents the M packages in the package database and the historical package Each meal attribute corresponds to a similarity score, which is an M-dimensional vector. The size is also dimensional vector, , express Recommended scores for historical meals.
[0031] S3.5. Calculate the recommendation score F of each package in the package database based on the historical time. The calculation formula is:
[0032] in, is the weight of the corresponding time of the historical package. The closer the time of the historical package is to the current time, the more recent the passenger's taste and preference are, and the larger the weight value is. If the time of the historical package is farther away from the current time, the passenger's taste and preference are more likely to change, so the weight value is smaller. N represents the number of historical packages; S3.6. Sort the refund scores of all meal packages to obtain sorted meal information R. The sorting formula is:
[0033] in, The recommended scores of the packages are sorted from large to small, and the recommended scores of the packages involved in the sorting are all greater than 0.5. That is, only the packages with a recommended score greater than 0.5 in the package database are selected for sorting, and the packages with a recommended score less than 0.5 are not selected into the recommended meal information. Then, the mapping function The sorted meal information R is obtained, where R is the recommended meal information.
[0034] Specifically, F={ }, if the package with the highest recommendation score in F is , the second most recommended package is , then after sorting, .
[0035] S4, comprehensive ranking: Combine the fixed meal information and recommended meal information, sort and adjust the package p in the package database, and traverse all packages p. If ,and , then sort according to the package order of R, that is, the packages in the package database that belong to both R and P are sorted according to the package order of R, and the sorted set is ;like ,and , then according to That is, the packages in the package database that belong to R but not to P are sorted according to the package order of R. The sorted set is ;like ,and , then sort by the fixed order of meals, that is, the meals in the meal database that belong to P but not to R are sorted by the order of P meals, and the sorted set is , the final order of meal push is ={ }, and then push the packages in the package database to the passengers in the sorting order of T, that is, push them to the passenger's meal selection interface for passengers to choose. The packages with the highest sorting are more in line with the passengers' personal preferences, which makes it convenient for passengers to quickly and easily choose the packages they want and reduces the complexity of passengers' meal selection information.
[0036] It should be noted that the sorting method of T is a sorting method that prioritizes fixed meal information. If some packages belong to both the fixed meal information and the recommended meal information, then these packages are sorted first, and the packages are sorted according to the package order of the recommended meal information; if some packages only belong to the fixed meal information or only belong to the recommended meal information, then the packages belonging to the fixed meal information are sorted first, with the fixed packages being prioritized, and then the packages belonging only to the recommended meal information are sorted. The present invention integrates the recommended meal information and the fixed meal information for recommendation sorting, which not only meets the personalized needs of passengers, but also meets the recommendation needs of airlines for customized meal selection.
[0037] S5. Precise meal search: In addition to directly sorting the packages in the package database and pushing them to the passengers, the present invention also provides an intelligent search function for packages, that is, a search box is set in the meal selection interface, and the passenger enters the search information I in the search box to obtain the meal recommendation results. , the formula is as follows:
[0038] Among them, RNN is a sorting model, which obtains relevant packages according to the search information I, and sorts the packages according to the sorting method of T and pushes them to passengers. For example, there are beef fried rice packages and ham fried rice packages in the package database. After calculation, the order of these two packages is that the beef fried rice package is ranked first. After the passenger enters the search information of "beef" in the search box, since the ingredients of the beef fried rice package and the ham fried rice package both contain beef, the meal selection interface will push the push information of the beef fried rice package and the ham fried rice package, and the beef fried rice package is ranked first for passengers to choose. Therefore, the present invention adds an intelligent search function on the basis of the final sorting of the recommended packages, accurately captures the demand tendency of passengers, and thus tailors the package selection plan, so that passengers can experience convenience and personalized services in the meal ordering process.
[0039] The following further describes step S3 through a specific embodiment: S3.1. Obtain the historical package information h of the passenger. The historical package information includes two packages: ham fried rice and cured meat clay pot rice. The packages in the package database include three packages: beef fried rice, porridge and seafood noodles. Only the ingredient components of the meal content are used as attributes. The attributes of the beef fried rice package include beef, rice, egg, pepper and apple. The attributes of the ham fried rice include lunch beef, rice, corn and carrot. S3.2, substitute the historical meal information h and the meal information in the meal database S_meals into the formula for vectorization, and obtain H and S, where H is a vector matrix of K×u×N, S is a vector matrix of K×u×M, the number of N is 2, including two meal sets of ham fried rice and cured meat clay pot rice; the number of M is 3, including three meal sets of beef fried rice, porridge and seafood noodles. Taking beef fried rice as an example, the vectorized result is as follows:
[0040] Among them, the beef fried rice set has 6 attributes, and each attribute has multiple vector dimensions, while the ham fried rice has only four attributes. For the convenience of calculation, the vector dimensions of two attributes will be expanded to 0, that is, two attributes with vector dimensions of 0 are added for subsequent calculations.
[0041] S3.3. Obtain the feature similarity c between the meal attributes of each historical meal and each meal in the meal database through the Pearson correlation coefficient. Taking the ham fried rice meal as an example, the feature similarity between each attribute and each meal in the meal database is as follows:
[0042] S3.4. Obtain the similarity weight value corresponding to each meal attribute in the historical meal set, and calculate the recommendation score of each meal set in the meal set database and the corresponding historical meal set in terms of attributes. , also taking the ham fried rice set meal as an example, the recommended scores of each set meal in the set meal database and the ham fried rice set meal in terms of attributes are as follows:
[0043] The calculation process is as follows: The recommended score of the beef fried rice meal in terms of attributes is: 0.97×0.6+0.61×0.15+0.124×0.05+0.24×0.05=0.6917; The recommended score of the porridge meal in terms of attributes is: 0.34×0.6+0.725×0.15+0.35×0.05+0.26×0.05=0.3432; The recommended score for the seafood noodle set meal in terms of attributes is: 0.57×0.6+0.24×0.15+0.36×0.05+0.16×0.05=0.403.
[0044] S3.5. Calculate the recommendation score F of each set meal in the set meal database based on the historical time. The time weight of the ham fried rice set meal is 0.5, and the time weight of the cured meat clay pot rice is 0.2. The recommendation score of each set meal in the set meal database is as follows:
[0045] The calculation process is as follows: The recommended score for the beef fried rice set meal is: 0.6917×0.8+0.6015×0.5=0.8541; The recommended score for the porridge meal is: 0.3432×0.8+0.2174×0.5=0.3832; The recommended score for the seafood noodle set meal is: 0.403×0.8+0.3862×0.5=0.5155; S3.6. Sort F to obtain sorted meal information R, that is, the original order of F's meal packages is beef fried rice meal package, porridge meal package, and seafood mixed noodles meal package. After sorting according to the recommendation score, the order of R's meal packages is sorted from large to small according to the recommendation score as beef fried rice meal package, seafood mixed noodles meal package, and porridge meal package. However, the recommendation score of the porridge meal package is less than 0.5, so it does not participate in the sorting, that is, it is not selected into the recommended meal information. That is, the recommended meal information R includes beef fried rice meal package and seafood mixed noodles meal package.
[0046] The above description is only a preferred embodiment of the present invention and is not a limitation on the design of this case. Any equivalent changes made based on the design key of this case shall fall within the protection scope of this case.
Claims
1. A personalized recommendation method for flight meals, characterized in that: The following steps are involved: S1. Database construction: Establishing a meal package database and a passenger database, wherein the meal package database includes a plurality of meal packages and meal package labels, wherein the meal package labels include meal prices, applicable cabins for meals, applicable membership levels for meals, whether meals are applicable to point redemption activities and point redemption prices, meal names, meal pictures, meal descriptions, meal contents, and meal labels; The passenger database includes passenger user information, passenger itinerary information, and historical meal selection information. The passenger user information includes historical itinerary information, membership level, and points balance; the passenger itinerary information includes the passenger's current flight information and cabin information; the historical itinerary information includes the passenger's historical flight information and historical cabin information; and the historical meal selection information includes historical meal package information selected corresponding to the historical itinerary information; S2. Generate fixed meal information: The fixed meal information is a meal list specified for passengers of different membership levels, points balances and flight levels. Based on the passenger user information and passenger itinerary information, combined with the meal tag of the meal database, the database query and combined output of the fixed meal information P are performed. ; S3, generating recommended meal information, including the following steps: S3.
1. Obtain the passenger's historical meal information h, where both the historical meal information and the meal in the meal database contain multiple meal attributes; S3.2, vectorize the historical package information h and the package information in the package database S_meals, and the vectorization formula is: Among them, H is a vector matrix of K×u×N, S is a vector matrix of K×u×M, K represents the number of meal attributes, u represents the vector dimension size corresponding to a single meal attribute, N represents the number of historical packages, and M represents the number of packages in the package database; Represents a language model, which is used to vectorize package information; S3.
3. The feature similarity C between the historical packages and the packages in the package database is obtained through the Pearson correlation coefficient. The calculation formula is: C PCorrelation in, for vector matrix, K represents the number of meal attributes, M represents the number of meal packages in the meal package database, and N represents the number of historical meal packages. contains the feature similarity c between the meal attributes of each historical meal and each meal in the meal database, for The vector matrix of , Indicates shared Similarity matrix of historical meals S3.
4. Obtain the similarity weight value corresponding to each meal attribute in the historical package, and calculate the The total recommendation score of each package in the package database based on the attributes of the historical packages , the calculation formula is: in, Indicates The weight value corresponding to each meal attribute, K represents the number of meal attributes, Represents the M packages in the package database and the historical package Each meal attribute corresponds to a similarity score, which is an M-dimensional vector. The size is dimensional vector, , express Recommendation scores for historical meals; S3.
5. Calculate the recommendation score F of each package in the package database based on the historical time. The calculation formula is: in, is the weight of the corresponding time of the historical package. The closer the time of the historical package is to the current time, the greater the weight value. N represents the number of historical packages; S3.
6. Sort the recommendation scores of all meal packages to obtain sorted meal information R. The sorting formula is: in, The recommended scores of the packages are sorted from large to small, and the recommended scores of the packages involved in the sorting are all greater than 0.5, and then the mapping function is used Obtain sorted meal information R, where R is the recommended meal information; S4, comprehensive ranking: Combine the fixed meal information and recommended meal information to sort and adjust the package p in the package database, and traverse all packages p. If p ,and , then sort according to the order of R's packages, and the sorted set is ;like ,and , then according to The order of packages is sorted, and the sorted set is ;like ,and , then sort according to the order of P's packages, and the sorted set is , the final sorting order of meal push is ={ }, and then push the meal packages in the meal package database to the meal selection interface in the sorting order of T for passengers to choose; S5. Precise meal search: The meal selection interface has a search box. Passengers can enter search information I in the search box to obtain meal recommendation results. , the formula is as follows: Among them, RNN is a sorting model, which obtains relevant packages according to the retrieval information I, and sorts the packages according to the sorting method of T and pushes them to passengers.
2. A personalized flight meal recommendation method according to claim 1, characterized in that: In step S1, the meal content is the ingredients of the meal, and the meal label is a description of the type, taste, and dishes of the meal; the flight information includes the flight number, flight date, departure place, destination, and travel time period.
3. A personalized flight meal recommendation method according to claim 2, characterized in that: In step S3.1, the meal attributes include meal name, meal description, meal content, and meal label. The meal content is the ingredients of the meal set, and each ingredient is an attribute.
4. The method for personalized recommendation of flight meals according to claim 1, characterized in that: In step S3.6, F={ }, if the package with the highest recommendation score in F is , the second most recommended package is ,but .
5. The method for personalized recommendation of flight meals according to claim 1, characterized in that: In step 3.6, if the number of historical meal packages is 0, only the fixed meal information P is pushed to the passenger.