Dynamic flight itinerary-oriented recommendation model training method, recommendation method and device

Abstract

The invention belongs to the technical field of flight recommendation, and provides a dynamic flight itinerary-oriented recommendation model training method and device, and a dynamic flight itinerary-oriented recommendation method and device, and the method comprises the steps: obtaining information of flights historically purchased by passengers; the flight information comprises a plurality of attributes; splitting the attributes through a fine-grained attribute splitting method to obtain a data set, and dividing the data set into a training set and a test set; wherein the data set is a set of triple data; training the initialized recommendation model by using the training set, and evaluating the recommendation model by using the test set to obtain a trained recommendation model; according to the method, the attributes of the flight information are split through a fine-grained attribute splitting method, and complex attribute information can be effectively split and quantified, so that the accuracy and efficiency of flight prediction by a recommendation model are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of flight recommendation, and in particular relates to a recommendation model training method, a recommendation method and a device for dynamic flight itineraries. Background Art

[0002] With the continuous advancement of globalization and the improvement of people's living standards, air travel has become an indispensable part of daily life. The complexity of modern transportation is increasing day by day, and the amount of information users face when choosing flights has increased dramatically. How to find the most suitable itinerary plan from the massive amount of flight information has become an urgent problem to be solved. The current flight itinerary recommendation system on the market mainly relies on simple rule sorting strategies, such as sorting by price, departure time, punctuality, etc. However, this static rule-based sorting method is difficult to effectively match the diverse needs of users, resulting in poor recommendation results.

[0003] Traditional flight search systems mainly help users choose flights by price sorting, departure time sorting, and punctuality sorting. Price sorting arranges flights from low to high by price, allowing users to quickly find the cheapest flight. However, this method ignores the user's needs for other factors, such as departure time flexibility and airline service quality, and cannot fully meet user needs. In departure time sorting, the system sorts flights according to their departure time, allowing users to choose the flight that best suits their schedule, but it also ignores factors such as flight price and service quality. In punctuality sorting, the system sorts flights by their punctuality. Although it improves the reliability of flights, it cannot comprehensively consider the user's time and cost needs, and the recommendation results are not personalized and accurate enough.

[0004] Compared with traditional static recommendation products, such as books and movies, flight itinerary recommendations have obvious dynamic characteristics. Flight prices, flight policies, seat availability, etc. are all changing in real time, and these changes directly affect users' purchasing decisions. Traditional recommendation methods, such as matrix decomposition and item-based or user-based collaborative filtering, are difficult to cope with such dynamic changes. Specifically, the complexity of flight recommendations is mainly reflected in the following aspects:

[0005] Dynamic features: Flight prices, flight policies, seat availability and other features are dynamic. For example, the price of the same flight may change in different time periods, and the flight seat availability may also change at any time. These changes require the recommendation system to have the ability to update and adjust in real time, while existing collaborative filtering methods usually rely on historical behavior data and static features, making it difficult to quickly respond to these dynamic changes.

[0006] Combination flight problem: Flight itinerary recommendations include not only direct flights, but also a variety of flight combination options. For example, users can choose to complete their trip from one city to another through a transfer or a jointly operated flight. The complexity of this combination itinerary increases the difficulty of recommendation. Traditional recommendation algorithms find it difficult to effectively handle such diverse combination schemes because these schemes involve multi-segment connection planning, and the recommendation system needs to consider multiple factors to make decisions.

[0007] Information island problem: Flight itinerary information is often fragmented. For example, when a user searches for flights from Hangzhou to Singapore, it is difficult for the system to associate flights from Shanghai to London with a simple query. This information island feature makes it difficult for the existing network structure to effectively and uniformly process flight information, limiting the recommendation system in providing comprehensive recommendations. The information isolation between flights increases the difficulty of integrating different flight information in the recommendation system. Summary of the invention

[0008] In response to the problems in the background technology, the present invention proposes a recommendation model training method, a recommendation method and a device for dynamic flight itineraries, which can effectively integrate and process the dynamic characteristics of flights, combine flight plans and solve the problem of information islands.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] In a first aspect, the present disclosure proposes a recommendation model training method for dynamic flight itineraries, comprising:

[0011] Obtaining the flight information that the passenger has purchased historically; the flight information includes several attributes;

[0012] Splitting the attributes by a fine-grained attribute splitting method to obtain a data set, and dividing the data set into a training set and a test set; wherein the data set is a set of triple data;

[0013] The training set is used to train the initialized recommendation model, and the test set is used to evaluate the recommendation model to obtain a trained recommendation model.

[0014] Preferably, the recommendation model is a graph neural network.

[0015] Preferably, obtaining the flight information that the passenger has purchased in the past includes: obtaining the flight information that the passenger has purchased in the past through a user identifier.

[0016] Preferably, splitting the attributes by a fine-grained attribute splitting method includes:

[0017] filtering said attributes;

[0018] Classifying the attributes into categories;

[0019] Constructing a mapping relationship between the category and the number, and replacing the category with the number according to the mapping relationship;

[0020] Calculate the proportion of each category in the corresponding attribute to obtain the category proportion, and determine the scoring rule based on the category proportion;

[0021] Determine the score for each category according to the scoring rules;

[0022] According to the user identifier, the number and the score corresponding to the number, triple data is obtained.

[0023] Preferably, the use of the training set to train the initialized recommendation model includes:

[0024] Constructing a sparse matrix based on the training set;

[0025] The initialized recommendation model is trained based on the sparse matrix.

[0026] Preferably, initializing the recommendation model includes: initializing parameters of the recommendation model through uniform distribution.

[0027] Preferably, before splitting the attributes by a fine-grained attribute splitting method to obtain a data set, the method further includes: preprocessing the flight information.

[0028] In a second aspect, the present disclosure provides a recommendation model building device for dynamic flight itineraries, comprising:

[0029] A historical flight information acquisition module is used to acquire the flight information that the passenger has purchased historically; the flight information includes several attributes;

[0030] A fine-grained attribute splitting module, used to split the attributes by a fine-grained attribute splitting method, obtain a data set, and divide the data set into a training set and a test set; wherein the data set is a set of triple data;

[0031] The training module is used to train the initialized recommendation model using the training set and evaluate the recommendation model using the test set to obtain a trained recommendation model.

[0032] In a third aspect, the present disclosure proposes a recommendation method for dynamic flight itineraries, comprising:

[0033] Acquire a number of alternative flights; wherein the alternative flights include a number of attributes;

[0034] Input a number of alternative flights into the recommendation model trained in the first aspect to obtain recommended alternative flights.

[0035] In a fourth aspect, the present disclosure provides a recommendation device for dynamic flight itineraries, comprising:

[0036] The alternative flight acquisition module is used to acquire a number of alternative flights; wherein the alternative flights include a number of attributes;

[0037] The recommendation module is used to input a number of alternative flights into the recommendation model trained in the first aspect to obtain recommended alternative flights.

[0038] Beneficial effects of the present invention:

[0039] 1. The method of the present invention splits the attributes of flight information through a fine-grained attribute splitting method, which can effectively split and quantify complex attribute information, thereby improving the accuracy and efficiency of the recommendation model for flight prediction. This method can not only capture the user's preferences and behavior patterns more accurately, but also subdivide the flight information into specific attributes, such as departure time, flight duration, fare, etc., so that the recommendation model can better understand and match the user's personalized needs. Moreover, fine-grained attribute splitting also solves the problem of data sparsity and improves the robustness of the model in processing high-dimensional features and nonlinear relationships.

[0040] 2. By using user identification (such as a unique identifier such as a user ID or user name) to obtain the passenger's historically purchased flight information without collecting any other private information of the user, such as sensitive data such as name, age, gender or contact information, it avoids excessive collection of the user's detailed personal information and greatly reduces the risk of privacy leakage.

[0041] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A flow chart of a recommendation model training method for dynamic flight schedules of the present invention is shown;

[0044] Figure 2 A flow chart showing a method for recommending dynamic flight schedules according to the present invention;

[0045] Figure 3 (a) shows a flow chart of the fine-grained attribute splitting method of the present invention;

[0046] Figure 3 (b) is a schematic diagram showing the scoring rules according to the category proportion in the fine-grained attribute splitting method of the present invention;

[0047] Figure 4 The recommendation model training flow chart of the present invention is shown;

[0048] Figure 5 The recommended model architecture diagram of the present invention is shown;

[0049] Figure 6 The framework diagram of the device for building a recommendation model for dynamic flight schedules of the present invention is shown;

[0050] Figure 7 The framework diagram of the recommendation device for dynamic flight itineraries of the present invention is shown;

[0051] Figure 8 A schematic diagram of the device structure of the present invention is shown. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Reference Figure 1 As shown, a personalized recommendation method for dynamic flight itineraries is shown, which specifically includes the following steps:

[0054] S1. Obtain the flight information that the passenger has purchased in the past; the flight information includes several attributes;

[0055] Exemplarily, flight information includes the following attributes: airline, departure time, landing time, flight duration, fare, main cabin, etc.;

[0056] In the current flight recommendation system, privacy protection is also an important consideration. Traditional systems usually need to collect a large amount of user personal information, such as personal data, preference data, etc., to achieve the recommendation function. This method is prone to the risk of privacy leakage. A balance needs to be found between data processing and user privacy protection. Therefore, the flight information purchased by the passenger in the past can be obtained through the user ID;

[0057] By using user identification (such as a unique identifier such as a user ID or user name) to obtain the passenger's historically purchased flight information without collecting any other private information of the user, such as sensitive data such as name, age, gender or contact information, excessive collection of the user's detailed personal information is avoided, greatly reducing the risk of privacy leakage.

[0058] S2. Split the attributes by a fine-grained attribute splitting method to obtain a data set, and divide the data set into a training set and a test set; wherein the data set is a set of triple data;

[0059] As a preferred embodiment of the present disclosure, the training set and the test set are split in a ratio of 8:2.

[0060] Among them, refer to Figure 3 As shown in the figure, the attributes are split by the fine-grained attribute splitting method to obtain the data set, which specifically includes the following steps:

[0061] S21, filter attributes;

[0062] Specifically, flight attribute columns that have an impact on user preferences are selected. It is understandable that the number of attribute columns selected may not be fixed and may be selected based on existing information and empirical rules. For example, five attribute columns are selected: airline, take-off time, landing time, flight duration, fare, and main cabin.

[0063] S22, classify the attributes;

[0064] Specifically, for each selected attribute column, the number of categories is divided according to its characteristics and actual business needs. For example, for the airline attribute column, the types of airlines in the statistical data set will be counted, and each airline will be a category; for the take-off time and landing time, there will be six time period types based on every four hours, namely: 00:00-04:00, 04:00-08:00, 08:00-12:00, 12:00-16:00, 16:00-20:00, 20:00-24:00; for the flight duration attribute, it will be divided into four levels according to the duration, namely less than 2 hours, 2-4 hours, 4-6 hours, and more than 6 hours; for the fare attribute, it will also be divided into four categories according to the price, namely: less than 500 yuan, 500 to 1500 yuan, 1500 to 3000 yuan, and more than 3000 yuan; for the main cabin attribute, after counting the cabin types, four cabin categories are obtained.

[0065] S23, constructing a mapping relationship between categories and numbers;

[0066] Specifically, the categories are remapped to a unique number for each category, which is convenient for subsequent processing and calculation. For example, the first class is numbered 1, the business class is numbered 2, and the economy class is numbered 3.

[0067] S24, according to the mapping relationship, replace the category with the number;

[0068] Specifically, after the mapping relationship is constructed, the category is replaced with the number. For example, if the first-class cabin number is 1, the attribute value of the main cabin type being first-class cabin is replaced with 1, thereby obtaining data of all numerical types.

[0069] S25, calculating the proportion of each category in the corresponding attribute to obtain the category proportion;

[0070] Specifically, the calculation method of the category ratio is to divide the number of category purchases by the total number of purchases of the passenger, that is, category ratio = category purchases / total number of purchases. For example, for a user with a user ID of 10001, a total of 20 purchase records are found in the passenger's historical flight information, and 10 of the 20 purchase records are first-class. For user 10001, the first-class ratio is 50%.

[0071] S26. Determine the scoring rules based on the category proportions;

[0072] Specifically, a set of scoring rules is formulated based on the category proportion. This rule is used to convert the category proportion into a score. For example: when a category proportion is between 80% and 100%, the score of the category is 5 points; when the proportion is between 60% and 80%, the score is 4 points; when the proportion is between 40% and 60%, the score is 3 points; when the proportion is between 20% and 40%, the score is 2 points; when the proportion is between 0% and 20%, the score is 1 point.

[0073] S27. Determine the score of each category according to the scoring rules;

[0074] The categories are scored according to the proportions, and each category is scored according to the proportions of the categories grouped and counted by the user ID according to the scoring rules in step S26. For example, the first class accounts for 50% of the purchase records of user 10001, and according to the scoring rules, the first class is scored 3 points.

[0075] S28, obtaining triple data according to the user identifier, the number and the score corresponding to the number;

[0076] By splitting the attributes of flight information using the fine-grained attribute splitting method, complex attribute information can be effectively split and quantified, thereby improving the accuracy and efficiency of the recommendation model for flight prediction. This method can not only capture user preferences and behavior patterns more accurately, but also subdivide flight information into specific attributes such as departure time, flight duration, fare, etc., so that the recommendation model can better understand and match the user's personalized needs. Moreover, fine-grained attribute splitting also solves the problem of data sparsity and improves the robustness of the model when dealing with high-dimensional features and nonlinear relationships.

[0077] As a preferred embodiment of the present disclosure, since flight data is highly dynamic, it is necessary to clean and organize the flight information to remove invalid data and noise data. Therefore, before splitting the attributes by the fine-grained attribute splitting method and obtaining the data set, it also includes: preprocessing the flight information. Specifically, the flight information is cleaned and normalized.

[0078] S3, using the training set to train the initialized recommendation model, and using the test set to evaluate the recommendation model to obtain a trained recommendation model;

[0079] Reference Figure 5 As shown, specifically, the recommended model is a graph neural network. The data set in step S2 is used to train the model, wherein the training set is used to dynamically adjust parameters during model training, and the test set is used by users to evaluate the effect of the model. The training set and test set data are input into the graph neural network for training. During the training process, according to the fixed number of rounds T set by the user, the test module is called after each T round of training to evaluate the current model weight using the test set and observe the test indicators. When the maximum number of epochs set by the user is reached, the training is terminated, and the model weights are saved locally for prediction. During the training process, the loss function and Adam optimizer are used to adjust the parameters. The loss function is:.

[0080] When the maximum number of epochs set by the user is reached, the training is terminated and the model weights are saved locally for prediction. During the training process, the loss function and Adam optimizer are used to adjust the parameters. The loss function is: In the formula, is the normalization factor, and |σ| is the number of elements in the set σ.

[0081] In each iteration, the model makes predictions based on the current parameters and calculates the error between the predicted result and the actual result. The training ends when the indicator reaches the set threshold. In each layer of the graph convolution, the vector of the next layer is iteratively calculated. For user u, the embedding vector update formula of the k+1th layer is: The update formula for item i is: In each round of training, the model effect of that round is evaluated through the loss function, and the Adam optimizer is used to adjust the model parameters to minimize the loss function. The scoring calculation formula during the training process is: In the formula, e u is the final embedding vector of user u, e i is the final embedding vector of flight i. This inner product value is used to generate a ranking score to sort the recommendation results. In this example, the model is tested with the test set every 5 rounds of training and the evaluation index is calculated. The evaluation index uses the root mean square error to evaluate the degree of fit between the predicted score and the true score: In the formula, is the predicted rating of user u for item i, r u,i is the actual rating value, which represents the real rating or feedback of user u on item i. m is the total number of rating pairs, that is, the number of rating pairs in the test set. Each pair of ratings includes a predicted value and an actual value.

[0082] Reference Figure 4 As shown in the figure, the training set is used to train the initialized recommendation model, which specifically includes the following steps:

[0083] S311, constructing a sparse matrix based on the training set;

[0084] Sparse matrices are part of the processing of input data and are used to represent the relationship between users and flight attribute categories. Each row in the matrix represents a user, each column represents a flight attribute category, and the values ​​in the matrix represent the user's preference rating for a certain flight. Constructing sparse matrices helps improve computational efficiency and model performance. The user-item interaction matrix R is a matrix of size R. M×N where M is the number of users and N is the number of items. Based on the user-item interaction matrix R, we get the adjacency matrix of the user-item graph Therefore, the diagonal matrix D is a diagonal matrix of size (M+N)×(M+N), and each diagonal element D ii Represents the number of non-zero elements in the i-th row of the adjacency matrix A, also known as the degree of the node. The final embedding representation is used for model prediction and is expressed as a weighted sum of multiple levels of embedding: or In the formula, is a symmetric normalized adjacency matrix.

[0085] S312, training the recommendation model initialized based on the sparse matrix;

[0086] Sparse matrices are part of the processing of input data and are used to represent the relationship between users and flight attribute categories. Each row in the matrix represents a user, each column represents a flight attribute category, and the values ​​in the matrix represent the user's preference score for a certain flight. Constructing sparse matrices helps improve computational efficiency and model performance.

[0087] Initializing the recommendation model specifically includes: initializing the parameters of the recommendation model by uniform distribution. As a preferred embodiment of the present disclosure, the number of graph convolution layers of the recommendation model is set to 3 layers.

[0088] Reference Figure 6 As shown, based on the same inventive concept, the present disclosure also proposes a recommendation model building device for dynamic flight itineraries, including:

[0089] The historical flight information acquisition module is used to obtain the flight information that the passenger has purchased historically; the flight information includes several attributes;

[0090] A fine-grained attribute splitting module is used to split the attributes by a fine-grained attribute splitting method, obtain a data set, and divide the data set into a training set and a test set; wherein the data set is a set of triple data;

[0091] The training module is used to train the initialized recommendation model using the training set and evaluate the recommendation model using the test set to obtain a trained recommendation model.

[0092] Reference Figure 2 As shown, based on the same inventive concept, the present disclosure also proposes a recommendation method for dynamic flight itineraries, which specifically includes the following steps:

[0093] S4. Obtain several alternative flights; wherein the alternative flights include several attributes;

[0094] Obtain several alternative flight itineraries, use the trained recommendation model to make recommendations, and obtain recommended alternative flight itineraries;

[0095] S5. Input a number of alternative flights into the trained recommendation model to obtain recommended alternative flights.

[0096] Specifically, each alternative itinerary is divided into multiple attribute values, and the trained model is used to calculate the user's preference score for each attribute. The final prediction score is obtained by summing up the scores of each attribute, and the total score is calculated as follows: In the formula, α a is the weight of attribute a, is the predicted score of user u for attribute a.

[0097] Specifically, the score corresponding to each flight alternative itinerary is obtained, and the flight alternative itineraries are sorted according to the score.

[0098] Reference Figure 7 As shown, based on the same inventive concept, the present disclosure also proposes a recommendation device for dynamic flight itineraries, including:

[0099] The alternative flight acquisition module is used to acquire a number of alternative flights; wherein the alternative flights include a number of attributes;

[0100] The recommendation module is used to input a number of alternative flights into the above-mentioned trained recommendation model to obtain recommended alternative flights.

[0101] Reference Figure 8 As shown, based on the same inventive concept, the present disclosure also proposes a device, including a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and when the processor executes the computer instructions, the above-mentioned recommendation model training method for dynamic flight itineraries and / or the recommendation method for dynamic flight itineraries are executed.

[0102] Based on the same inventive concept, the present disclosure also proposes a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed, the above-mentioned recommendation model training method for dynamic flight itineraries and / or recommendation method for dynamic flight itineraries can be implemented.

[0103] Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0104] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0105] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A recommendation model training method for dynamic flight schedules, characterized in that: include: Obtaining the flight information that the passenger has purchased historically; the flight information includes several attributes; Splitting the attributes by a fine-grained attribute splitting method to obtain a data set, and dividing the data set into a training set and a test set; wherein the data set is a set of triple data; The training set is used to train the initialized recommendation model, and the test set is used to evaluate the recommendation model to obtain a trained recommendation model.

2. The recommendation model training method for dynamic flight schedules according to claim 1, characterized in that: The recommendation model is a graph neural network.

3. The recommendation model training method for dynamic flight schedules according to claim 2, characterized in that: Acquiring the flight information that the passenger has purchased in the past includes: acquiring the flight information that the passenger has purchased in the past through the user ID.

4. The recommendation model training method for dynamic flight schedules according to claim 3, characterized in that: The attributes are split by a fine-grained attribute splitting method, including: filtering said attributes; Classifying the attributes into categories; Constructing a mapping relationship between the category and the number, and replacing the category with the number according to the mapping relationship; Calculate the proportion of each category in the corresponding attribute to obtain the category proportion, and determine the scoring rule based on the category proportion; Determine the score for each category according to the scoring rules; According to the user identifier, the number and the score corresponding to the number, triple data is obtained.

5. The recommendation model training method for dynamic flight schedules according to claim 2, characterized in that: The using the training set to train the initialized recommendation model includes: Constructing a sparse matrix based on the training set; The initialized recommendation model is trained based on the sparse matrix.

6. The recommendation model training method for dynamic flight schedules according to claim 5, characterized in that: Initializing the recommendation model includes: initializing parameters of the recommendation model through uniform distribution.

7. The recommendation model training method for dynamic flight schedules according to claim 2, characterized in that: Before splitting the attributes by a fine-grained attribute splitting method to obtain a data set, the method further includes: preprocessing the flight information.

8. A device for building a recommendation model for dynamic flight schedules, characterized in that: include: A historical flight information acquisition module is used to acquire the flight information that the passenger has purchased historically; the flight information includes several attributes; A fine-grained attribute splitting module, used to split the attributes by a fine-grained attribute splitting method, obtain a data set, and divide the data set into a training set and a test set; wherein the data set is a set of triple data; The training module is used to train the initialized recommendation model using the training set and evaluate the recommendation model using the test set to obtain a trained recommendation model.

9. A recommendation method for dynamic flight itineraries, characterized in that: include: Acquire a number of alternative flights; wherein the alternative flights include a number of attributes; Input a plurality of alternative flights into the trained recommendation model described in any one of claims 1 to 7 to obtain recommended alternative flights.

10. A recommendation device for dynamic flight itineraries, characterized in that: include: The alternative flight acquisition module is used to acquire a number of alternative flights; wherein the alternative flights include a number of attributes; A recommendation module is used to input a number of alternative flights into the trained recommendation model described in any one of claims 1-7 to obtain recommended alternative flights.

Images