Ticketing Recommendation Method, System, Travel Server, and Program

Abstract

An embodiment of the present application provides a ticket recommendation method, system, travel server, and program. The method includes: obtaining a ticket search request of a user; determining search results of recommended transportation tickets and multiple candidate alternative search information of alternative transportation tickets; for any candidate alternative search information, determining a search preference score of the user for the candidate alternative search information; and for any candidate alternative search information, determining a score of the feature difference corresponding to each ticket feature of the candidate alternative search information compared with the recommended transportation ticket based on multiple ticket features related to the user's ticket decision-making, and obtaining the feature difference score of the candidate alternative search information; determining the score of each candidate alternative search information according to the search preference score and the feature difference score of each candidate alternative search information; and determining the recommended alternative search information according to the scores of each candidate alternative search information. The embodiment of the present application can improve the recommendation accuracy of alternative transportation ticket recommendations.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of computer Internet, and particularly to a ticket recommendation method, system, travel server and program. Background Art

[0002] The travel service platform is an online platform that provides travel services for users, such as OTP (Online Travel Platform), OTA (Online Travel Agency) platforms, etc. With the continuous development of intelligent technologies, the travel service platform not only provides ticket services such as querying, booking, changing, refunding and rescheduling of transportation tickets such as air tickets and train tickets, but also provides diversified travel solutions such as hotel reservation, travel packages, car rental, etc.

[0003] As an important part of the ticket service of the travel service platform, ticket recommendation is mainly used to recommend search results of recommended transportation tickets such as air tickets and train tickets that meet the ticket search conditions based on ticket search conditions such as the departure place, destination and departure date submitted by the user. Further, to meet the diverse travel needs and preferences of users, in addition to recommending search results of recommended transportation tickets that meet the ticket search conditions, ticket recommendation also performs alternative transportation ticket recommendation.

[0004] Alternative transportation ticket recommendation refers to recommending alternative transportation tickets that are not completely consistent with but related to the ticket search conditions for the user, so as to provide more choice possibilities for the user's travel, thereby helping the user find suitable transportation tickets such as air tickets and train tickets. The recommendation accuracy of alternative transportation ticket recommendation is crucial for improving the user's travel experience and increasing the ticket click-through rate and conversion rate of the travel service platform. Therefore, how to improve the recommendation accuracy of alternative transportation ticket recommendation has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0005] In view of this, the embodiments of the present application provide a ticket recommendation method, system, travel server and program to improve the recommendation accuracy of alternative transportation ticket recommendation.

[0006] To achieve the above object, the embodiments of the present application provide the following technical solutions.

[0007] In a first aspect, the embodiments of the present application provide a ticket recommendation method, including:

[0008] Obtaining a ticket search request of a user, where the ticket search request includes ticket search conditions;

[0009] Determining search results of recommended transportation tickets that meet the ticket search conditions;

[0010] Determine multiple candidate alternative search information for alternative transportation tickets related to the ticket search conditions. Each candidate alternative search information includes ticket search conditions for searching alternative transportation tickets and is associated with multiple ticket features related to the user's ticket decision-making;

[0011] For any candidate alternative search information, determine the search preference score of the user for the candidate alternative search information according to the user's user feature information, historical behavior information, the ticket search conditions of the candidate alternative search information, and the search results of the recommended transportation tickets;

[0012] Moreover, for any candidate alternative search information, based on the multiple ticket features, determine the score of the feature difference corresponding to each ticket feature of the candidate alternative search information compared with the recommended transportation tickets, and based on the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information, obtain the feature difference score of the candidate alternative search information;

[0013] According to the search preference scores of the user for each candidate alternative search information and the feature difference scores of each candidate alternative search information, determine the scores of each candidate alternative search information;

[0014] According to the scores of each candidate alternative search information, determine the recommended alternative search information from the multiple candidate alternative search information.

[0015] In a second aspect, an embodiment of the present application provides a ticket recommendation system, including:

[0016] A request acquisition module, configured to acquire a ticket search request of a user, where the ticket search request includes ticket search conditions;

[0017] A ticket search engine, configured to determine the search results of recommended transportation tickets that meet the ticket search conditions;

[0018] An alternative transportation ticket recommendation subsystem, configured to:

[0019] Determine multiple candidate alternative search information for alternative transportation tickets related to the ticket search conditions. Each piece of candidate alternative search information includes the ticket search conditions for searching alternative transportation tickets and is associated with multiple ticket characteristics related to the user's ticket decision-making. For any piece of candidate alternative search information, determine the search preference score of the user for the candidate alternative search information based on the user's characteristic information, historical behavior information, the ticket search conditions of the candidate alternative search information, and the search results of the recommended transportation tickets. And, for any piece of candidate alternative search information, based on the multiple ticket characteristics, determine the score of the characteristic difference corresponding to each ticket characteristic of the candidate alternative search information compared to the recommended transportation ticket, and based on the scores of the characteristic differences corresponding to each ticket characteristic of the candidate alternative search information, obtain the characteristic difference score of the candidate alternative search information. Determine the scores of each piece of candidate alternative search information according to the search preference scores of the user for each piece of candidate alternative search information and the characteristic difference scores of each piece of candidate alternative search information. Determine the recommended alternative search information from the multiple pieces of candidate alternative search information according to the scores of each piece of candidate alternative search information.

[0020] In a third aspect, an embodiment of the present application provides a travel server, including a memory and a processor. The memory stores computer execution instructions, and the processor calls the computer execution instructions to execute the ticket recommendation method as described in the first aspect above.

[0021] In a fourth aspect, an embodiment of the present application provides a computer program product, including computer execution instructions, which when executed, implement the ticket recommendation method as described in the first aspect above.

[0022] The embodiment of the present application combines the search preference score of the user for the candidate alternative search information and the characteristic difference score of the candidate alternative search information, can comprehensively consider the user's search preference and the actual situation of the alternative transportation tickets corresponding to the candidate alternative search information, generate accurate scores for the candidate alternative search information, and then this dual evaluation system of the search preference score and the characteristic difference score for the candidate alternative search information can reduce the error caused by one-sided consideration of a single factor, more accurately recommend alternative search information for alternative transportation tickets to the user, and improve the recommendation accuracy of alternative transportation ticket recommendations; thus providing more choice possibilities for the user's travel, enhancing the user's travel experience, and then increasing the ticket click-through rate and conversion rate of the travel service platform. Description of the Drawings

[0023] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0024] Figure 1 It is a page example diagram of the travel service platform.

[0025] Figure 2 It is a flowchart of the ticket recommendation method provided by the embodiment of the present application.

[0026] Figure 3 It is a flowchart of determining the search preference score provided by the embodiment of the present application.

[0027] Figure 4 It is a block diagram of the alternative transportation ticket recommendation subsystem provided by the embodiment of the present application.

[0028] Figure 5 It is a block diagram of the main network module provided by the embodiment of the present application.

[0029] Figure 6 It is a flowchart of determining the feature difference score provided by the embodiment of the present application.

[0030] Figure 7 It is a block diagram of the monotonic balance network module provided by the embodiment of the present application.

[0031] Figure 8 It is a block diagram of the ticket recommendation system provided by the embodiment of the present application. Detailed implementation manners

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of them. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0033] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties. And the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0034] To facilitate understanding of alternative transportation ticket recommendations, Figure 1 An example of a page example of a travel service platform is shown in FIG. Figure 1 As shown, the search page is a page provided by the travel service platform for users to search for air tickets, train tickets and other transportation tickets; when the user selects a mode of transportation (such as airplanes, trains and other modes of transportation), the search page includes a search box for users to enter ticket search conditions such as ODD, where ODD is the abbreviation of Origin, Destination and Departure Date; specifically, the departure place is the starting city or place of the user's trip, such as the departure place corresponds to the search field of depcity, the destination is the end city or place of the user's trip, such as the destination corresponds to the search field of arrcity, and the departure date is the date of the user's trip, such as the departure date corresponds to the search field of deptime.

[0035] For ease of explanation, the departure place when the user searches for tickets is called the target departure place, the destination when the user searches for tickets is called the target destination, and the departure date when the user searches for tickets is called the target departure date.

[0036] Users can submit ticket search requests to the travel service platform through the search page. The ticket search request can indicate the user's ticket search conditions such as the target departure point, target destination and target departure date when the mode of transportation is selected; further, the travel service platform can search and recommend search results of recommended transportation tickets that meet the ticket search conditions based on the user's ticket search conditions (for example, when the mode of transportation is selected, search and recommend search results of recommended transportation tickets that meet the target departure point, target destination and target departure date); at the same time, in order to meet users' diverse travel needs and travel preferences, the travel service platform also recommends alternative transportation tickets to users through alternative transportation ticket recommendations, which are not completely consistent with the ticket search conditions but are related and meet user needs (such as price requirements, distance requirements, etc.), thereby searching and recommending search results of alternative transportation tickets through the alternative search information of alternative transportation tickets. For example, if the search results for recommended transportation tickets that meet the ticket search conditions show that there are no remaining tickets or are not preferred by the user, the alternative search information of the alternative transportation tickets recommended by the alternative transportation ticket recommendation can provide a further transportation ticket search entry for the user's travel, thereby providing the user with more options for travel, helping the user find suitable transportation tickets, thereby improving the user's travel experience and increasing the ticket click-through rate and conversion rate of the travel service platform.

[0037] Combination Figure 1As shown, the Listing Page is a page provided by the travel service platform to display search results for users after they conduct ticket searches on the travel service platform. The Listing Page can display search results of recommended transportation tickets that match the user's ticket search criteria, as well as alternative search information for alternative transportation tickets recommended by the alternative transportation ticket recommendation. Specifically, the search results of recommended transportation tickets can be one or more items, such as one or more search results of recommended transportation tickets that match the target departure location, target destination, and target departure date. Among them, one search result of recommended transportation tickets includes information such as the departure time (deptime), arrival time (arrtime), price, discount, and other attribute information of a transportation schedule (direct transportation schedule or transfer transportation schedule) that matches the target departure location, target destination, and target departure date.

[0038] It should be noted that transportation schedules, such as flight schedules of airplanes, train schedules of trains (such as high-speed rail schedules of high-speed rails), bus schedules of buses, etc., depend on the specific ticket scenarios and selected transportation modes when users conduct ticket searches, and the embodiments of this application do not have limitations. Other attribute information is additional attribute information related to transportation schedules (such as flight schedules). Taking flight schedules as an example, other attribute information includes, for example, the class of the flight, the airline, and whether it is a direct flight.

[0039] The alternative transportation ticket recommendation can be a multi-mode recommendation, mainly used to recommend alternative search information for alternative transportation tickets corresponding to multiple alternative recommendation modes. Among them, different alternative recommendation modes indicate different alternative transportation ticket scenarios. For the convenience of distinction and description, the search information used to search for alternative transportation tickets is called alternative search information. The alternative search information for alternative transportation tickets can at least include the ticket search criteria for alternative transportation tickets. Specifically, the alternative search information for alternative transportation tickets can include ticket search criteria such as the departure location, destination, departure date, and selected transportation mode under the corresponding alternative recommendation mode. Thus, through the alternative search information for alternative transportation tickets, search results of alternative transportation tickets that match the departure location, destination, departure date, and transportation mode under the corresponding alternative recommendation mode can be searched.

[0040] For example, in combination with Figure 1As shown, multiple alternative recommendation modes for alternative transportation tickets can include, but are not limited to, the near-departure mode, the intermodal mode (such as the air-rail intermodal mode that combines air and train intermodal services), other direct transportation modes (such as the high-speed rail direct mode in the air ticket search scenario), etc.; among them, the alternative transportation ticket recommendation in the near-departure mode can also be called the near-ticket recommendation, which recommends alternative search information for alternative transportation tickets based on the proximity of the departure place, destination, or departure date, so as to search for the corresponding search results of transportation tickets for the near departure place, near destination, or near departure date; further, the near-departure mode can be specifically divided into the near OD (departure place / destination) mode and the near-date mode.

[0041] The near OD mode can also be called the near-location mode, which mainly recommends alternative search information for alternative transportation tickets based on the proximity of the departure place and / or destination, and belongs to the near-ticket recommendation in the alternative transportation ticket recommendation; specifically, the near OD mode corresponds to one or more alternative search information for alternative transportation tickets, which is used to recommend alternative transportation tickets for the near departure place adjacent to the target departure place and / or the near destination adjacent to the target destination when the target departure date remains unchanged, so as to search for the search results of alternative transportation tickets that match the target departure date, near departure place, and / or near destination under the selected transportation mode.

[0042] In the near OD mode, the alternative search information for alternative transportation tickets can include information such as the departure place (target departure place or near departure place), destination (target destination or near destination), departure date (target departure date), and price, and the transportation mode is the transportation mode selected by the user when searching for tickets on the search page; among them, the departure place in the alternative search information for alternative transportation tickets in the near OD mode is the near departure place, and / or the destination is the near destination, and the price is the price of the transportation ticket that matches the departure place, destination, and departure date in the alternative search information, such as the minimum price.

[0043] It should be noted that based on the specific transportation mode, the near departure place can be regarded as a city or location with transportation stations such as airplanes or trains within the preset geographical range of the target departure place, and the number of near departure places may be one or more; similarly, the near destination can be regarded as a city or location with transportation stations such as airplanes or trains within the preset geographical range of the target destination, and the number of near destinations may be one or more. By way of example, in the case of air tickets, the cities or locations with airports within the geographical range with the target departure place as the center and a preset radius distance are the near departure places, and the cities or locations with airports within the geographical range with the target destination as the center and a preset radius distance are the near destinations.

[0044] The near - date mode mainly recommends alternative search information for alternative transportation tickets based on the proximity of the departure date, belonging to the near - ticket recommendation in alternative transportation ticket recommendation. Specifically, the near - date mode corresponds to one or more alternative search information for alternative transportation tickets, which is used to recommend alternative search information for alternative transportation tickets with near - departure dates close to the target departure date when the target departure place and target destination remain unchanged, so as to search for search results of alternative transportation tickets that match the target departure place, target destination, and near - departure date under the selected mode of transportation. For example, the near - departure date close to the target departure date is, for instance, a date within a preset number of days from the target departure date. The near - departure date can be one or more, depending on the specific number of days of the preset number of days. In the near - date mode, the alternative search information for alternative transportation tickets can include information such as the departure place (target departure place), destination (target destination), departure date (near - departure date), and price, and the mode of transportation is the mode of transportation when the user conducts ticket search on the search page. Among them, the price is the price of the transportation ticket that matches the departure place, destination, and departure date in the alternative search information, such as the minimum price.

[0045] The inter - modal transportation mode (such as the air - train inter - modal transportation mode) corresponds to one or more alternative search information for alternative transportation tickets, which is used to recommend alternative search information for alternative transportation tickets combined with multiple means of transportation when the target departure place, target destination, and target departure date remain unchanged, so as to search for search results of multiple means of transportation combined that match the target departure place, target destination, and target departure date under the mode of transportation combined with multiple means of transportation. That is to say, the inter - modal transportation mode is an alternative recommendation mode that combines different types of means of transportation (such as airplanes, trains, buses, etc.) to recommend ticket combinations of multiple means of transportation that meet the travel needs of the target departure place, target destination, and target departure date to users. In the inter - modal transportation mode, the alternative search information for alternative transportation tickets can include information such as the departure place (target departure place), destination (target destination), departure date (target departure date), inter - modal transportation vehicle identifier (the inter - modal transportation vehicle identifier is used to identify the multiple means of transportation combined, such as the identifiers of the airplane and high - speed rail in the air - train inter - modal transportation mode), and price. Among them, the price is the price of the transportation ticket combination of multiple means of transportation combined that matches the departure place, destination, and departure date in the alternative search information, such as the minimum price.

[0046] Other direct transportation modes (such as the high-speed rail direct mode in the flight ticket search scenario), corresponding to one or more alternative search information for alternative transportation tickets, are used to recommend alternative search information for alternative transportation tickets of other direct transportation under the condition that the target departure place, target destination, and target departure date remain unchanged, so as to search for search results of alternative transportation tickets that match the target departure place, target destination, and target departure date under other direct transportation modes. For example, in the flight ticket search scenario, if there is a direct high-speed rail between the target departure place and the target destination, the corresponding alternative search information for alternative transportation tickets can be recommended through the high-speed rail direct mode to search for search results of direct high-speed rails that match the target departure place, target destination, and target departure date. Under other direct transportation modes, the alternative search information for alternative transportation tickets can include information such as the departure place (target departure place), destination (target destination), departure date (target departure date), direct transportation tool identifier (such as the identifier of the high-speed rail transportation tool in the high-speed rail direct mode), and price. Among them, the price is the price of the transportation tickets of other direct transportation that match the departure place, destination, and departure date in the alternative search information, such as the minimum price.

[0047] The alternative search information for alternative transportation tickets can include the ticket search conditions for alternative transportation tickets, which are used to search for search results of transportation tickets that match the departure place, destination, departure date, and corresponding transportation mode of the alternative search information. For example, after the user selects the alternative search information for a certain alternative transportation ticket on the search result page, the departure place, destination, departure date, and transportation mode of the alternative search information for alternative transportation tickets can be used as further ticket search conditions, so that the travel service platform can enter a further search result page. This further search result page can display search results of transportation tickets that match the departure place, destination, departure date, and transportation mode of the alternative search information selected by the user, providing more ticket selection possibilities for the user's travel. The above-mentioned further search result page will not only display search results of transportation tickets that match the ticket search conditions, but also further display alternative search information for alternative transportation tickets corresponding to various alternative recommendation modes through alternative transportation ticket recommendation, so as to continuously provide the user with further search opportunities for alternative transportation tickets on the search result page. That is to say, the alternative transportation ticket recommendation is not limited to being triggered by the user's ticket search through the search page or by the alternative search information for alternative transportation tickets displayed on the search result page.

[0048] Further, the search results of recommended transportation tickets that match the user's ticket search conditions may be blank information. For example, there may be no transportation stations corresponding to the transportation mode selected by the user at the target departure place and / or the target destination. As a result, there are no search results that match the user's ticket search conditions on the search result page returned by the travel service platform to the user. At this time, alternative transportation ticket recommendations (especially the nearby ticket recommendations in alternative transportation ticket recommendations) can recommend alternative search information for alternative transportation tickets near the departure place and / or near the destination, so as to provide the user with a further search entry for transportation tickets near the departure place and / or near the destination to meet the user's travel needs.

[0049] It can be seen that the alternative search information of alternative transportation tickets, as the search entry for searching alternative transportation tickets displayed in the alternative transportation ticket recommendation on the search result page, can provide the user with a further search opportunity for alternative transportation tickets when the search results of recommended transportation tickets that match the ticket search conditions displayed on the search result page show no remaining tickets, or are not liked by the user, or are blank information. Thus, it helps the user find suitable transportation tickets to meet the user's travel needs, and further improves the user's travel experience and the ticket click-through rate and conversion rate of the travel service platform. Therefore, alternative transportation ticket recommendation is an important part of the ticket recommendation of the travel service platform, and it has important technical significance to improve the recommendation accuracy of alternative transportation ticket recommendation.

[0050] Based on this, the embodiments of the present application provide a ticket recommendation solution applied to a travel service platform to improve the recommendation accuracy of alternative transportation ticket recommendation, thereby enhancing the user's travel experience and increasing the ticket click-through rate and conversion rate of the travel service platform.

[0051] As an optional implementation, Figure 2 An optional flowchart of the ticket recommendation method provided by the embodiments of the present application is exemplarily shown. The ticket recommendation method can be applied to a travel service platform, such as an OTP, OTA platform, etc. In the optional implementation, the travel service platform (such as an OTP, OTA platform) may include a server on the service side, referred to as a travel server (such as an OTP server, OTA server). Thus, the ticket recommendation method provided by the embodiments of the present application can be specifically applied to the travel server. In the optional implementation, the travel service platform (such as an OTP, OTA platform) can be regarded as an online service platform formed by a server cluster formed by travel servers through setting various software and hardware functional architectures.

[0052] Referring to Figure 2 , the ticket recommendation method provided by the embodiments of the present application may include the following steps.

[0053] Step S210: Obtain a ticket search request of a user, where the ticket search request includes ticket search conditions.

[0054] In an alternative implementation, the ticket search request can be triggered through a search page. For example, when the user selects a mode of transportation, the user can input and submit ticket search conditions such as the target departure location, target destination, and target departure date through the search page. Then, the user's client device can transmit the ticket search request corresponding to the ticket search conditions to the travel service platform for ticket search.

[0055] In an alternative implementation, the ticket search request can also be triggered by alternative search information for alternative transportation tickets. In the embodiments of the present application, the alternative search information for alternative transportation tickets includes ticket search conditions for alternative transportation tickets and can be used to search for alternative transportation tickets. Therefore, the embodiments of the present application also support triggering of the user's ticket search request by the alternative search information for alternative transportation tickets. For example, if the user clicks on the alternative search information for a certain alternative transportation ticket displayed on the search result page, the user's client device can transmit the ticket search request corresponding to the alternative search information to the travel service platform.

[0056] As an alternative implementation, the ticket search conditions include, but are not limited to, the target departure location, target destination, target departure date, etc. Further, the ticket search conditions can also indicate the mode of transportation, etc.

[0057] Exemplarily, if the ticket search request is triggered through the search page, the mode of transportation can be determined in advance. For example, before the user's client device enters the search page, the user has pre-selected search topics such as flight ticket search or train ticket search. Thus, the mode of transportation such as an airplane or a train when the user's client device enters the search page can be determined in advance. Furthermore, the departure location entered by the user on the search page is the target departure location in the ticket search conditions, the destination entered by the user on the search page is the target destination in the ticket search conditions, and the departure date entered by the user on the search page is the target departure date in the ticket search conditions.

[0058] Exemplarily, if the ticket search request is triggered by alternative search information for alternative transportation tickets, the departure location included in the alternative search information is the target departure location in the ticket search conditions, the destination included in the alternative search information is the target destination in the ticket search conditions, and the departure date included in the alternative search information is the target departure date in the ticket search conditions. Further, if the alternative search information for alternative transportation tickets also includes an identifier for the mode of transportation, such as the identifier for the intermodal transportation means in the alternative search information corresponding to the intermodal mode, the identifier for the non-stop transportation means in the alternative search information corresponding to other non-stop transportation modes, etc., then the mode of transportation corresponding to the identifier for the mode of transportation in the alternative search information is the mode of transportation for the ticket search conditions; if the alternative search information does not include an identifier for the mode of transportation, then the mode of transportation for the ticket search conditions can be the mode of transportation when the user conducts a ticket search on the search page. For example, the mode of transportation for the alternative search information corresponding to the near OD mode and the near date mode is the mode of transportation when the user conducts a ticket search on the search page.

[0059] Step S220: Determine the search results of recommended transportation tickets that match the ticket search conditions.

[0060] After obtaining the ticket search request, the travel service platform can conduct a ticket search to search for and recommend the search results of recommended transportation tickets that match the ticket search conditions. For the convenience of explanation, the transportation tickets that match the ticket search conditions searched and recommended by the travel service platform are called recommended transportation tickets. Step S220 can be regarded as the normal ticket recommendation of the travel service platform. In an optional implementation example, the travel service platform can set up a ticket search engine to search for and recommend the search results of recommended transportation tickets that match the ticket search conditions based on the ticket search conditions.

[0061] Step S230: Determine multiple candidate alternative search information for alternative transportation tickets related to the ticket search conditions. Each candidate alternative search information includes the ticket search conditions for searching alternative transportation tickets and is associated with multiple ticket characteristics related to the user's ticket decision-making.

[0062] In the embodiments of the present application, in addition to normal ticket recommendations, the travel service platform also conducts alternative transportation ticket recommendations to recommend alternative search information for alternative transportation tickets that are not completely consistent with the ticket search conditions but are relevant to the user, so as to search for alternative transportation tickets. In an optional implementation, the alternative transportation ticket recommendation can be divided into a recall stage and a scoring and recommendation stage; in the recall stage, the embodiments of the present application can first preliminarily recall multiple candidate alternative search information of alternative transportation tickets that are not completely consistent with the ticket search conditions but are relevant to the ticket search conditions. For the convenience of description, the candidate alternative search information of alternative transportation ticket candidates is called candidate alternative search information; in the scoring and recommendation stage, the embodiments of the present application can score each candidate alternative search information, and then determine the finally recommended alternative search information based on the scores of each candidate alternative search information.

[0063] Step S230 can be regarded as the recall stage of the alternative transportation ticket recommendation, which is used to determine multiple candidate alternative search information of alternative transportation tickets that are not completely consistent with the ticket search conditions but are relevant to the ticket search conditions.

[0064] In an optional implementation, the alternative transportation ticket recommendation can have multiple alternative recommendation modes, including but not limited to the nearby location mode (i.e., the nearby OD mode), the nearby date mode, the intermodal mode (such as the air-rail intermodal mode), and other direct transportation modes (such as the high-speed rail direct mode), etc. The embodiments of the present application can determine the candidate alternative search information of alternative transportation tickets corresponding to various alternative recommendation modes, so as to obtain multiple candidate alternative search information of alternative transportation tickets. Among them, each alternative recommendation mode corresponds to at least one candidate alternative search information, depending on the specific situation and definition of the alternative recommendation mode.

[0065] That is to say, the candidate alternative search information of alternative transportation tickets corresponding to each alternative recommendation mode is determined based on different logics and rules. Specifically, by combining the ticket search conditions with each alternative recommendation mode, the candidate alternative search information of alternative transportation tickets related to the ticket search conditions under each alternative recommendation mode is determined to recall multiple candidate alternative search information of alternative transportation tickets.

[0066] For example, taking the nearby location mode as an example, the embodiments of the present application can set a preset geographical range for the target departure place and the target destination, so as to screen out the nearby departure places and nearby destinations with transportation stations (transportation stations corresponding to the transportation mode of the ticket search conditions) through the set preset geographical range; thus, based on the unchanged target departure date, when the departure place is switched to the nearby departure place and / or the destination is switched to the nearby destination, at least one candidate alternative search information of alternative transportation tickets corresponding to the nearby location mode is determined.

[0067] Exemplarily, taking the near - date mode as an example, embodiments of the present application can set a date range close to the departure date. For example, the date within a preset number of days from the target departure date is the near - departure date. Thus, based on the unchanged mode of transportation, target departure place, and target destination in the ticket search conditions, when the departure date is switched to each near - departure date respectively, at least one piece of candidate alternative search information for alternative transportation tickets corresponding to the near - date mode is determined.

[0068] Exemplarily, taking the inter - modal transportation mode as an example, embodiments of the present application can set multiple means of transportation that need to be combined. Thus, based on the unchanged target departure place, target destination, and target departure date, when the mode of transportation is adjusted to a combination of multiple means of transportation in the inter - modal transportation mode, at least one piece of candidate alternative search information for alternative transportation tickets corresponding to the inter - modal transportation mode is determined.

[0069] Exemplarily, taking other direct - transportation modes as an example, embodiments of the present application can set the mode of transportation for other direct transportation. Thus, based on the unchanged target departure place, target destination, and target departure date, when the mode of transportation is adjusted to the mode of transportation for other direct transportation, at least one piece of candidate alternative search information for alternative transportation tickets corresponding to other direct - transportation modes is determined.

[0070] After determining multiple pieces of candidate alternative search information for alternative transportation tickets related to the ticket search conditions, each piece of candidate alternative search information can include ticket search conditions for searching alternative transportation tickets, such as ticket search conditions for the departure place, destination, departure date, etc. for searching alternative transportation tickets; further, the ticket search conditions of the candidate alternative search information can also include the mode of transportation corresponding to the alternative transportation tickets, such as the identification of the means of transportation, etc.

[0071] In embodiments of the present application, the candidate alternative search information for alternative transportation tickets not only includes the ticket search conditions for alternative transportation tickets to be used for searching alternative transportation tickets, but also is associated with multiple ticket features related to the user's ticket decision. Among them, the ticket features related to the user's ticket decision can be understood as ticket feature information that affects the user's decision - making when the user selects transportation tickets, and is used to further score each piece of candidate alternative search information in the scoring and recommendation stage after recalling multiple pieces of candidate alternative search information.

[0072] In an optional implementation, the multiple ticket features related to the user's ticket decision associated with the candidate alternative search information for alternative transportation tickets can include, but are not limited to:

[0073] Price feature, which represents the price of the alternative transportation ticket corresponding to the candidate alternative search information for subsequent analysis of the price difference between the alternative transportation ticket and the recommended transportation ticket; price is one of the decision-making factors for users when choosing transportation tickets, and there may be a negative correlation between price and the tendency of users to choose transportation tickets. For example, the lower the price of a transportation ticket, the higher the tendency of users to choose it; specifically, compared with the recommended transportation ticket, as the search entry for the alternative transportation ticket, if the price of the alternative transportation ticket is lower than that of the recommended transportation ticket, the tendency of users to choose the alternative transportation ticket is higher. Therefore, each alternative search information can be associated with a price feature to recommend, as much as possible, an alternative transportation ticket with a lower price than the recommended transportation ticket; in an alternative implementation, the price feature associated with the candidate alternative search information can be determined based on the ticket search conditions included in the candidate alternative search information. For example, based on the ticket search conditions such as the departure place, destination, departure date, and transportation mode of the candidate alternative search information, the price (such as the minimum price) of the matching transportation ticket is determined as the price feature associated with the candidate alternative search information;

[0074] Distance feature, which represents the distance between the departure location of the alternative transportation ticket corresponding to the candidate alternative search information and the target departure location, and the distance between the destination of the alternative transportation ticket corresponding to the candidate alternative search information and the target destination. For example, the comprehensive distance between the departure location of the alternative transportation ticket and the target departure location, and the distance between the destination of the alternative transportation ticket and the target destination; the distance feature associated with the candidate alternative search information is used to analyze the distance difference from the distance feature of the recommended transportation ticket in the subsequent process. Among them, the distance feature of the recommended transportation ticket is represented as the distance between the departure location of the recommended transportation ticket and the target departure location, and the distance between the destination and the target destination, such as the comprehensive distance; the distance feature is one of the decision-making factors for users when choosing transportation tickets. Specifically, when the departure location of the alternative transportation ticket is a nearby departure location and / or the destination is a nearby destination, the distance feature can be used to calculate the distance difference from the distance feature of the recommended transportation ticket, so as to help determine whether the alternative transportation ticket can provide a convenient travel option for users when the departure location of the alternative transportation ticket is a nearby departure location and / or the destination is a nearby destination. For example, the alternative transportation ticket with a smaller distance difference can provide a convenient travel option for users, and thus has a higher tendency to be selected by users; by way of example, in the nearby location mode, users will expect the recommended nearby departure location to be closer to the departure location of the recommended transportation ticket (such as the target departure location), and the recommended nearby destination to be closer to the destination of the recommended transportation ticket (such as the target destination); in an alternative implementation, the distance feature associated with the candidate alternative search information can be determined based on the ticket search conditions included in the candidate alternative search information. For example, based on the ticket search conditions such as the departure location and destination of the candidate alternative search information, determine the distance between the departure location of the candidate alternative search information and the target departure location, and the distance between the destination of the candidate alternative search information and the target destination, as the distance feature associated with the candidate alternative search information;

[0075] Time feature, which represents the departure time of the alternative transportation ticket corresponding to the candidate alternative search information, such as the departure date and departure time point of the alternative transportation ticket, for use in analyzing the time difference between the departure time of the alternative transportation ticket and the departure time of the recommended transportation ticket in the subsequent process. The departure time of the recommended transportation ticket is, for example, the departure date and departure time point of the recommended transportation ticket; the time feature is one of the decision-making factors for users when choosing transportation tickets. Specifically, in the nearby date mode, the departure time of the alternative transportation ticket can be used to calculate the time difference from the departure time of the recommended transportation ticket, so as to help the travel service platform preferentially recommend alternative transportation tickets with a smaller time difference; in an alternative implementation, the time feature associated with the candidate alternative search information can be determined based on the ticket search conditions included in the candidate alternative search information. For example, based on the ticket search conditions such as the departure location, destination, and departure date of the candidate alternative search information, determine the departure time (such as the earliest departure time) of the transportation ticket that matches, as the time feature associated with the candidate alternative search information.

[0076] It should be noted that the ticket features associated with the candidate alternative search information may be related to but not exactly the same as the ticket search conditions included in the candidate alternative search information. Specifically, the ticket search conditions included in the candidate alternative search information are mainly used to search for alternative transportation tickets, while the ticket features associated with the candidate alternative search information are the optimization and adjustment factors for determining the final recommended alternative search information in the embodiments of the present application, affecting the priority of the final recommendation. That is to say, whether a certain candidate alternative search information will be finally recommended needs to consider ticket features such as price features (used to determine the price difference), distance features (used to determine the distance difference), and time features (used to determine the time difference).

[0077] Step S240: For any candidate alternative search information, determine the search preference score of the user for the candidate alternative search information according to the user's user feature information, historical behavior information, the ticket search conditions of the candidate alternative search information, and the search results of the recommended transportation tickets.

[0078] After completing step S230, the embodiments of the present application can recall multiple candidate alternative search information, thus entering the scoring and recommendation stage from step S240 to step S270. By determining the scores of each candidate alternative search information, the alternative search information is finally recommended to improve the recommendation accuracy of the alternative transportation ticket recommendation.

[0079] In the embodiments of the present application, the score of the candidate alternative search information is mainly composed of two parts, namely the search preference score of the user for the candidate alternative search information and the feature difference score of the candidate alternative search information. Among them, the search preference score of the user for the candidate alternative search information can be regarded as the preference score of the user for selecting the candidate alternative search information for ticket search. For example, the preference score of the user for selecting the departure place, destination, and departure date of the candidate alternative search information for ticket search. At the level of multiple ticket features related to the user's ticket decision-making, the feature difference score of the candidate alternative search information can be regarded as the comprehensive score of the feature differences of the candidate alternative search information and the recommended transportation tickets in each ticket feature. Thus, the feature difference score of the candidate alternative search information, combined with the search preference score of the user for the candidate alternative search information, can obtain the score of the candidate alternative search information, which is used to determine the recommended alternative search information from multiple candidate alternative search information. Among them, the recommended alternative search information and the search results of the recommended transportation tickets are displayed together on the search result page.

[0080] Specifically, the search preference score of the candidate alternative search information for the user can reflect the preference degree of the user for selecting the candidate alternative search information when conducting ticket searches, that is, the preference of the user for the ticket search conditions (such as departure location, destination, departure date) included in the candidate alternative search information, so as to quantify the search preference for the candidate alternative search information. The travel service platform can preferentially recommend the candidate alternative search information with a higher search preference score for the user. At the same time, the feature difference score of the candidate alternative search information takes into account the differences in ticket features between the candidate alternative search information and the recommended transportation tickets, such as differences in price, distance, and time. The above differences can effectively measure the deviation degree of the alternative transportation tickets corresponding to the candidate alternative search information from the recommended transportation tickets in multiple dimensions of ticket features, thereby helping the travel service platform screen and recommend the alternative transportation tickets that are closest to the user's needs and improving the recommendation accuracy of the alternative transportation tickets. That is to say, the embodiments of the present application combine the search preference score of the candidate alternative search information for the user and the feature difference score of the candidate alternative search information, which can comprehensively consider the user's search preference and the actual situation of the alternative transportation tickets corresponding to the candidate alternative search information, generate an accurate score for the candidate alternative search information. Furthermore, this dual evaluation system of the search preference score and the feature difference score for the candidate alternative search information can reduce the errors caused by one-sided consideration of a single factor and more accurately match the user's preference and recommend alternative search information close to the user's needs.

[0081] In the embodiments of the present application, step S240 is used to determine the search preference score of the candidate alternative search information for the user during the scoring and recommendation stage of the alternative transportation ticket recommendation. Specifically, for any candidate alternative search information, the embodiments of the present application can determine the search preference score of the candidate alternative search information for the user based on the user's user characteristic information, historical behavior information, the ticket search conditions included in the candidate alternative search information, and the search results of the recommended transportation tickets. Among them, the input information on which the determination of the search preference score of the candidate alternative search information for the user depends may include:

[0082] The user's user characteristic information. The user's user characteristic information can be multiple items, including multiple user basic information of the user, and further can include multiple personalized information of the user. By way of example, multiple user basic information of the user such as the user's age, gender, occupation, location, etc., and multiple personalized information of the user such as the user's travel frequency, budget preference, etc. The user's user characteristic information can help the travel service platform understand the basic needs and habits of the user, and then make personalized adjustments to the recommendation results of the alternative transportation tickets.

[0083] The historical behavior information of the user. There can be multiple pieces of the user's historical behavior information. For example, multiple pieces of historical behavior information represent multiple behavior records of the user on the travel service platform, including but not limited to multiple behavior records related to transportation tickets on the travel service platform, such as historical search for transportation tickets, historical purchase of transportation tickets, historical browsing of transportation tickets, historical modification and refund of transportation tickets, etc. Each piece of historical behavior information (such as each behavior record) can include multiple historical behavior characteristics for recording the historical behavior, such as the time when the historical behavior occurred, the type of behavior, the details of the transportation tickets involved, etc. The historical behavior information of the user can help the travel service platform analyze the user's preferences and trends for transportation tickets.

[0084] The ticket search conditions of the candidate alternative search information. There can be multiple ticket search conditions for the candidate alternative search information. For example, the departure location, destination, departure date, etc. used to search for alternative transportation tickets in the candidate alternative search information.

[0085] The search results of the recommended transportation tickets, that is, the search results of the recommended transportation tickets that match the ticket search conditions based on the user's ticket search request by the travel platform service. The recommended transportation tickets are the transportation tickets screened by the travel service platform according to the recommendation strategy to meet the user's ticket search needs. The recommended transportation tickets not only consider the user's current ticket search conditions but can also further optimize the recommendation results based on the user's historical behavior and preferences. In the embodiments of this application, the search results of the recommended transportation tickets and the recommended alternative search information can be displayed together on the search result page. Thus, the search results of the recommended transportation tickets can be used as the context information of the to-be-recommended alternative search information to determine the search preference score of the user for the candidate alternative search information. As the context information of the to-be-recommended alternative search information, the search results of the recommended transportation tickets can be one or more, and each search result includes information such as the departure location, destination, and other attribute information.

[0086] It should be noted that the information used to determine the search preference score described above may not include ticket features related to the user's ticket decision, such as ticket features like price feature, distance feature, time feature, etc.; for example, the user feature information used to determine the search preference score described above does not include ticket features such as price feature, distance feature, time feature, etc.; the historical behavior information of the user used to determine the search preference score described above does not include ticket features such as price feature, distance feature, time feature, etc.; the ticket search conditions of the candidate alternative search information used to determine the search preference score described above do not include ticket features such as price feature, distance feature, time feature, etc.; the search results of the recommended transportation tickets used to determine the search preference score described above do not include ticket features such as price feature, distance feature, time feature, etc. It should be further noted that in the embodiment of the present application, when determining the feature difference score of the candidate alternative search information, ticket features such as price feature, distance feature, time feature, etc. related to the user's ticket decision will be used, while when determining the search preference score, the user feature information, historical behavior information, ticket search conditions of the candidate alternative search information, and search results of the recommended transportation tickets other than ticket features such as price feature, distance feature, time feature, etc. are used.

[0087] As an optional implementation example, Figure 3 Exemplarily shows an optional flowchart for determining the search preference score provided by the embodiment of the present application. Referring to Figure 3 , the optional process for the embodiment of the present application to determine the search preference score may include the following steps.

[0088] Step S310: Embed each piece of user feature information of the user into corresponding user feature vectors, and splice the user feature vectors into a user feature vector combination.

[0089] The process of embedding processing is to map discrete and high-dimensional features into a low-dimensional vector space; specifically, the user feature information of the user is multi-dimensional and includes multiple discrete information, such as multiple user feature information like the user's gender, age, personalized preferences, etc. Among them, each piece of user feature information can be respectively converted into a low-dimensional vector through embedding processing to obtain each user feature vector corresponding to each piece of user feature information; that is to say, each dimension of the user's user feature information (i.e., each piece of user feature information of the user) will be respectively converted into a low-dimensional vector to obtain the corresponding user feature vector. Furthermore, in the embodiment of the present application, each user feature vector can be spliced into a user feature vector combination, that is, each user feature vector is spliced and integrated into a comprehensive vector representation for subsequent processing and analysis.

[0090] In an optional implementation, each piece of user feature information of the user processed in step S310 may not include ticket features related to the user's ticket decision, such as price feature, distance feature, time feature, etc.

[0091] In an alternative implementation, Figure 4 An exemplary block diagram of an alternative transportation ticket recommendation subsystem provided by an embodiment of the present application is shown, in combination with Figure 4 As shown, the alternative transportation ticket recommendation subsystem 400 serves as the alternative transportation ticket recommendation part of the ticket recommendation system of the travel service platform. The alternative transportation ticket recommendation subsystem 400 may include:

[0092] A recall module 410, configured to determine multiple pieces of candidate alternative search information of alternative transportation tickets related to the ticket search conditions of the ticket search request;

[0093] A main network module 420, and the main network module 420 is configured to implement step S240, that is, the main network module 420 may, for any piece of candidate alternative search information, determine the search preference score of the user for the candidate alternative search information according to the user's user characteristic information, historical behavior information, the ticket search conditions of the candidate alternative search information, and the search results of the recommended transportation tickets.

[0094] As an alternative implementation, Figure 5 An exemplary optional block diagram of the main network module provided by an embodiment of the present application is shown, in combination with Figure 4 and Figure 5 As shown, the main network module 420 may include:

[0095] A main network data input module 510, configured to obtain the user's user characteristic information, historical behavior information, the ticket search conditions of each piece of candidate alternative search information (such as the departure place, destination, departure date), and the search results of the recommended transportation tickets; in an alternative implementation, the information obtained by the main network data input module 510 may not include ticket characteristics related to the user's ticket decision-making, such as price characteristics, distance characteristics, and time characteristics;

[0096] A basic network module 520, configured to determine the search preference score of the user for each piece of candidate alternative search information based on the user characteristic information, historical behavior information, the ticket search conditions of each piece of candidate alternative search information, and the search results of the recommended transportation tickets obtained by the main network data input module 510.

[0097] Specifically, the basic network module 520 may include a user characteristic processing module 521 that implements step S310, that is, the user characteristic processing module 521 may be configured to embed each piece of the user's user characteristic information (excluding ticket characteristics such as price characteristics, distance characteristics, and time characteristics) into corresponding user characteristic vectors, and splice the user characteristic vectors into a user characteristic vector combination.

[0098] Exemplarily, in combination with Figure 5As shown, the user feature processing module 521 may include multiple Embedding processing units and a splicing unit. In the user feature processing module 521, each Embedding processing unit is configured to respectively embed each piece of user feature information of the user into a corresponding user feature vector to obtain each user feature vector. In the user feature processing module 521, the splicing unit is configured to splice each user feature vector into a user feature vector combination.

[0099] Step S320: For any piece of historical behavior information of the user, embed each historical behavior feature in the historical behavior information into a corresponding historical behavior feature vector, and splice each historical behavior feature vector into a historical behavior feature vector combination to obtain the historical behavior feature vector combination corresponding to the historical behavior information.

[0100] As an alternative implementation, each piece of historical behavior information of the user can respectively undergo embedding processing and splicing processing to form a corresponding historical behavior feature vector combination, so that each piece of historical behavior information can respectively correspond to a historical behavior feature vector combination. Specifically, for any piece of historical behavior information of the user, the embodiments of the present application can perform embedding processing on each historical behavior feature in the historical behavior information to obtain corresponding historical behavior feature vectors, and then splice each historical behavior feature vector together to form the historical behavior feature vector combination corresponding to the historical behavior information. By forming a corresponding historical behavior feature vector combination for each piece of historical behavior information of the user, the user can be accurately profiled by analyzing each piece of historical behavior information.

[0101] In the alternative implementation, in combination with Figure 5 As shown, the basic network module 520 may include a historical behavior processing module 522 that implements step S320, that is, the historical behavior processing module 522 can, for any piece of historical behavior information of the user, embed each historical behavior feature in the historical behavior information into a corresponding historical behavior feature vector, and splice each historical behavior feature vector into a historical behavior feature vector combination to obtain the historical behavior feature vector combination corresponding to the historical behavior information.

[0102] Exemplarily, in combination with Figure 5As shown, the historical behavior processing module 522 may include multiple groups of embedding processing units and multiple splicing units. Among them, one group of embedding processing units may include multiple embedding processing units for performing embedding processing on a piece of historical behavior information, and one splicing unit is used for splicing the historical behavior feature vectors after the embedding processing of a piece of historical behavior information. Specifically, in the historical behavior processing module 522, one group of embedding processing units is used to respectively embed each historical behavior feature in a piece of historical behavior information of a user into corresponding historical behavior feature vectors; in the historical behavior processing module 522, one splicing unit is used to splice the historical behavior feature vectors after the embedding processing of a piece of historical behavior information into a corresponding historical behavior feature vector combination.

[0103] Step S330: Based on the historical behavior feature vector combinations corresponding to each piece of historical behavior information, extract the potential relationships of the historical behavior information through the self-attention mechanism and convert them into the historical behavior representation of the user.

[0104] As an optional implementation, after obtaining the historical behavior feature vector combinations corresponding to each piece of historical behavior information in the embodiments of the present application, the historical behavior feature vector combinations corresponding to each piece of historical behavior information can be input into the Transformer Layer for processing, so as to, through the self-attention mechanism of the Transformer Layer, extract the potential relationships of the historical behavior information of the user based on the historical behavior feature vector combinations corresponding to each piece of historical behavior information, and convert the potential relationships of the historical behavior information into the historical behavior representation of the user for subsequent recommendation or analysis.

[0105] It should be noted that the Transformer Layer is a neural network layer based on the self-attention mechanism, which is used to process sequence data, especially applicable in natural language processing and time series analysis. The core idea of the Transformer is to capture the relationships between various elements in the sequence through the self-attention mechanism, thereby enhancing the model's representation ability. In an alternative implementation, the Transformer Layer may include a multi-head attention layer and a feed-forward neural network layer; specifically, after the combined input of the historical behavior feature vectors corresponding to each piece of historical behavior information into the Transformer Layer, the Transformer Layer calculates the mutual relationships between the combined historical behavior feature vectors corresponding to different historical behavior information through the self-attention mechanism to understand the historical behavior information that affects the user's preferences and decisions; moreover, the Transformer Layer can simultaneously focus on different parts of the combined historical behavior feature vectors corresponding to multiple pieces of historical behavior information according to the multi-head attention, so as to obtain a rich context representation to identify the patterns and changes in the user's historical behavior; furthermore, the Transformer Layer outputs a comprehensive vector that combines multiple combined historical behavior feature vectors, that is, the historical behavior representation of the user.

[0106] In an alternative implementation, in combination with Figure 5 As shown, the basic network module 520 may include a transformer layer 523 that implements step S330, that is, the transformer layer 523 can extract the potential relationships of the historical behavior information based on the combined historical behavior feature vectors corresponding to each piece of historical behavior information and convert them into the historical behavior representation of the user.

[0107] Step S340: Embed each ticket search condition of the candidate alternative search information into the corresponding search condition vector, and splice the search condition vectors into the corresponding search condition vector combination.

[0108] As an alternative implementation, the ticket search conditions of the candidate alternative search information processed in step S340 do not include ticket features related to the user's ticket decision-making, such as price features, distance features, and time features. For example, the ticket search conditions of the candidate alternative search information processed in step S340 may include: the departure location, destination, and departure date of the candidate alternative search information, that is, the ODD of the candidate alternative search information; each ticket search condition of the above candidate alternative search information, as discrete features, can be mapped into a vector space through embedding processing, so that each ticket search condition of the candidate alternative search information is transformed into a corresponding search condition vector, that is, the vector representation of the ticket search condition. After each ticket search condition is converted into a corresponding search condition vector, each search condition vector can be concatenated into a long vector, that is, the search condition vector combination, representing the comprehensive feature vector of the ticket search conditions of the candidate alternative search information.

[0109] In an alternative implementation, as shown in Figure 5 the basic network module 520 may include a search condition processing module 524 that implements step S340, that is, the search condition processing module 524 can embed each ticket search condition of the candidate alternative search information into a corresponding search condition vector and concatenate each search condition vector into a corresponding search condition vector combination.

[0110] For example, as shown in Figure 5 the search condition processing module 524 may include multiple embedding processing units and a concatenation unit; in the search condition processing module 524, each embedding processing unit is used to embed each ticket search condition of the candidate alternative search information into a corresponding search condition vector respectively to obtain each search condition vector; in the search condition processing module 524, the concatenation unit is used to concatenate each search condition vector into a corresponding search condition vector combination.

[0111] Step S350: Embed each search result of the recommended transportation ticket into a corresponding search result vector and concatenate each search result vector into a corresponding search result vector combination.

[0112] In the embodiments of the present application, the search results of the recommended transportation tickets, as the context information of the candidate alternative search information to be recommended, participate in the confirmation of the search preference score of the candidate alternative search information. As an alternative implementation, the search results of the recommended transportation tickets processed in step S350 do not include ticket features related to the user's ticket decision-making, such as price features, distance features, and time features. The embodiments of the present application can embed each search result of the recommended transportation ticket except for ticket features such as price features, distance features, and time features into a corresponding search result vector, and then concatenate each search result vector into a corresponding search result vector combination.

[0113] In an alternative implementation, in combination with Figure 5 As shown, the basic network module 520 may include a search result processing module 525 for implementing step S350. That is, the search result processing module 525 may embed and process each search result of the recommended transportation tickets into a corresponding search result vector, and splice each search result vector into a corresponding search result vector combination.

[0114] Exemplarily, in combination with Figure 5 As shown, the search result processing module 525 may include a plurality of embedding processing units and a splicing unit; in the search result processing module 525, each embedding processing unit is used to embed and process each search result of the recommended transportation tickets into a corresponding search result vector; in the search result processing module 525, the splicing unit is used to splice each search result vector into a corresponding search result vector combination.

[0115] Step S360: Based on the historical behavior representation of the user and the search condition vector combination, through the self-attention mechanism, determine a comprehensive representation vector of the ticket search condition that fuses the user's historical behavior and the candidate alternative search information.

[0116] As an alternative implementation, after obtaining the historical behavior representation of the user through step S330 and obtaining the search condition vector combination through step S340 in the embodiments of the present application, the historical behavior representation of the user and the search condition vector combination may be input into an Attention Layer for processing, so as to extract the potential relationship between the user's historical behavior and the ticket search condition of the candidate alternative search information through the self-attention mechanism, and obtain a comprehensive representation vector of the ticket search condition that fuses the user's historical behavior and the candidate alternative search information.

[0117] It should be noted that the Attention Layer calculates the correlation or dependence between different inputs through the self-attention mechanism, and dynamically assigns weights to each input to represent the importance of the input in the current context. Specifically, the representation of the user's historical behavior can reflect the user's historical behavior and capture the user's interests and preferences; the search condition vector combination can represent the conditional features related to the alternative transportation ticket search of the candidate alternative search information (such as the departure place, destination, departure date, etc.). Thus, in the Attention Layer, the Attention Layer can analyze the correlation between the historical behavior representation and the search condition vector combination, determine the user behavior that has a strong impact on the ticket search conditions of the candidate alternative search information, and assign higher weights; furthermore, after being processed by the Attention Layer, the user's historical behavior representation and the search condition vector combination will obtain a new weighted representation, which synthesizes the relationship between the user's historical behavior and the ticket search conditions of the candidate alternative search information, that is, the comprehensive representation vector.

[0118] In an alternative implementation, in combination with Figure 5 As shown, the basic network module 520 may include an attention layer 526 that implements step S360, that is, the attention layer 526 can determine a comprehensive representation vector that integrates the user's historical behavior and the ticket search conditions of the candidate alternative search information based on the user's historical behavior representation and the search condition vector combination through the self-attention mechanism.

[0119] Step S370: Combine the user feature vector combination, the comprehensive representation vector, the search condition vector combination, and the search result vector combination into a combined vector; perform multi-layer feature mapping and non-linear processing on the combined vector to obtain the search preference score of the user for the candidate alternative search information.

[0120] After obtaining the user feature vector combination through step S310, the search condition vector combination through step S340, the search result vector combination through step S350, and the comprehensive representation vector that integrates the user's historical behavior and the ticket search conditions of the candidate alternative search information through step S360, the embodiments of the present application can combine the user feature vector combination, the comprehensive representation vector, the search condition vector combination, and the search result vector combination into a combined vector.

[0121] In an alternative implementation, the embodiments of the present application can input the combined vector into multiple fully connected (FC) layers and non-linear activation functions (such as the ReLU activation function or other non-linear activation functions) for processing, so as to perform multi-layer feature mapping and non-linear processing on the combined vector to transform and enhance the combined vector, and then output the search preference score of the user for the candidate alternative search information.

[0122] In an alternative implementation, in combination with Figure 5 As shown, the basic network module 520 may include a merging module 527 for implementing step S370, a combination 528 of multiple fully connected layers and a non-linear activation function; wherein, the merging module 527 is used to merge the user feature vector combination, the comprehensive representation vector, the search condition vector combination, and the search result vector combination into a merged vector; the combination 528 of multiple fully connected layers and a non-linear activation function is used to obtain the search preference score of the user for the candidate alternative search information through multi-layer feature mapping and non-linear processing of the merged vector.

[0123] It should be noted that the fully connected layer can map the input to a new feature space through a linear transformation, so as to integrate different inputs and capture the relationships and interactions between the inputs; in the embodiments of the present application, multiple fully connected layers can be set, and a non-linear activation function is connected after each fully connected layer. The non-linear activation function is, for example, a ReLU (Rectified Linear Unit) activation function. By introducing non-linearity through the non-linear activation function, it is possible to learn more complex patterns and relationships, rather than just performing a linear transformation. For example, the ReLU activation function can change each negative value to zero while keeping positive values unchanged, thereby helping to fit the relationship between complex inputs and outputs.

[0124] In an alternative implementation, the embodiments of the present application adopt a combination of multiple fully connected layers and a non-linear activation function, such as a stack of multiple fully connected layers and a non-linear activation function, for extracting higher-level feature representations layer by layer; specifically, after the merged vector undergoes feature mapping through the first fully connected layer, non-linear processing is performed through the non-linear activation function after the first fully connected layer; then, it undergoes feature mapping through the second fully connected layer, and then non-linear processing is performed through the non-linear activation function after the second fully connected layer, and so on, until the combination processing of multiple fully connected layers and a non-linear activation function is completed, and the search preference score of the user for the candidate alternative search information is output. That is to say, a non-linear activation function (such as a ReLU activation function) is connected after each fully connected layer to perform a non-linear transformation on the output of the fully connected layer. Furthermore, the stack of fully connected layers connected to non-linear activation functions in each layer can enable each layer to learn more abstract features based on the previous layer. Thus, through multi-layer feature mapping and non-linear processing, the representation ability of the merged vector is gradually improved, and finally the search preference score of the user for the candidate alternative search information is output.

[0125] Each candidate alternative search information respectively passes through Figure 3For the processing of the shown process, search preference scores for each candidate alternative search information can be obtained. Among them, the search preference score of the user for the candidate alternative search information can be regarded as a score output by the alternative transportation ticket recommendation subsystem of the travel service platform using the model, denoted as Logit. 1 ; where Logit is a term in machine learning, especially used in logistic regression models or binary classification tasks, representing the predicted value. In the embodiments of the present application, the Logit of the candidate alternative search information 1 The higher the score, the higher the degree of preference of the user for the candidate alternative search information.

[0126] Returning to Figure 2 As shown, in step S250, for any candidate alternative search information, based on multiple ticket features related to the user's ticket decision-making, the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information compared to the recommended transportation ticket are determined, and based on the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information, the feature difference score of the candidate alternative search information is obtained.

[0127] In the embodiments of the present application, step S250 is used to determine the feature difference score of the candidate alternative search information in the scoring and recommendation stage of alternative transportation ticket recommendation. Specifically, for any candidate alternative search information, the embodiments of the present application can determine the differences between the candidate alternative search information and the recommended transportation ticket in each ticket feature (such as price feature, distance feature, time feature), so as to obtain the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information compared to the recommended transportation ticket, simply referred to as the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information, and then comprehensively combine the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information to obtain the feature difference score of the candidate alternative search information; the feature difference score of the candidate alternative search information can reflect the differences between the candidate alternative search information and the recommended transportation ticket in multiple ticket features, thereby improving the accuracy of recommending alternative search information based on the feature difference score of the candidate alternative search information.

[0128] As an optional implementation example, Figure 6 Exemplarily shows an optional flowchart for determining the feature difference score provided by the embodiments of the present application. Referring to Figure 6 , the optional process for determining the feature difference score in the embodiments of the present application may include the following steps.

[0129] Step S610: Pool the multiple ticket features respectively associated with the search results of the recommended transportation ticket to obtain multiple pooled ticket features of the recommended transportation ticket.

[0130] The search results of recommended transportation tickets are used as context information for alternative search information to be recommended. Each of the search results of recommended transportation tickets may be associated with multiple ticket features related to the user's ticket decision. For example, each of the search results of recommended transportation tickets is respectively associated with a price feature, a distance feature, a time feature, etc. The relevant concepts of the price feature, the distance feature, and the time feature can be referred to the previous description and will not be elaborated here.

[0131] In the embodiments of the present application, multiple ticket features respectively associated with the search results of recommended transportation tickets can be subjected to pooling processing, so as to determine pooling ticket features that meet the pooling requirements in each feature dimension, and obtain multiple pooling ticket features of the recommended transportation tickets. That is, the multiple pooling ticket features of the recommended transportation tickets can include the ticket features that meet the pooling requirements in each feature dimension of the recommended transportation tickets, and are used to compare with the ticket features of the candidate alternative search information in each feature dimension. It should be noted that the pooling processing is a feature dimension reduction operation used to summarize and reduce the dimension of a group of features in any feature dimension. For example, the min pooling processing can select the feature with the smallest feature value from a group of features in any feature dimension.

[0132] That is to say, the ticket features related to the user's ticket decision have multiple feature dimensions, such as the price dimension of the price feature, the distance dimension of the distance feature, the time dimension of the time feature, etc. By performing pooling processing on the multiple ticket features respectively associated with the search results of recommended transportation tickets, the pooling ticket features with feature values that meet the pooling requirements can be selected from each feature dimension, and the pooling ticket features of the recommended transportation tickets in each feature dimension can be obtained. For example, the pooling price feature of the recommended transportation tickets in the price dimension, the pooling distance feature in the distance dimension, and the pooling time feature in the time dimension form multiple pooling ticket features of the recommended transportation tickets.

[0133] For example, taking the min pooling operation as an example, by performing min pooling processing on the multiple ticket features respectively associated with the search results of recommended transportation tickets, the pooling ticket features with the smallest feature values can be selected from each feature dimension. For example, the pooling price feature with the smallest price among the search results of recommended transportation tickets, the pooling distance feature with the shortest distance, and the pooling time feature with the smallest time form multiple pooling ticket features of the recommended transportation tickets.

[0134] In an alternative implementation, in combination with Figure 4As shown, the alternative traffic ticket recommendation subsystem 400 may include a monotonic balance network module 430. The monotonic balance network module 430 is used to implement step S250, that is, for any candidate alternative search information, the monotonic balance network module 430 can determine the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information compared to the recommended traffic ticket based on multiple ticket features related to the user's ticket decision, and obtain the feature difference score of the candidate alternative search information based on the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information.

[0135] As an optional implementation, Figure 7 An optional block diagram of the monotonic balance network module provided by the embodiments of the present application is exemplarily shown. In combination with Figure 4 and Figure 7 As shown, the monotonic balance network module 430 may include:

[0136] A monotonic network data input module 710, which is used to obtain multiple ticket features respectively associated with each search result of the recommended traffic ticket, and multiple ticket features associated with each candidate alternative search information;

[0137] A data preprocessing module 720. The data preprocessing module 720 is used to implement step S610, that is, the data preprocessing module 720 can perform pooling processing on multiple ticket features respectively associated with each search result of the recommended traffic ticket to obtain multiple pooled ticket features of the recommended traffic ticket.

[0138] In an optional implementation, the data preprocessing module 720 may be a pooling module. The pooling module takes multiple ticket features respectively associated with each search result of the recommended traffic ticket as input and outputs pooled ticket features that meet the pooling requirements for each feature dimension, that is, multiple pooled ticket features of the recommended traffic ticket; for example, the pooling module may be a minimum pooling module. The minimum pooling module takes multiple ticket features respectively associated with each search result of the recommended traffic ticket as input and outputs pooled ticket features with the minimum feature value for each feature dimension, that is, multiple pooled ticket features of the recommended traffic ticket.

[0139] Step S620: Perform difference processing on each ticket feature of the candidate alternative search information and each pooled ticket feature of the recommended traffic ticket respectively to obtain the feature difference of each ticket feature of the candidate alternative search information.

[0140] After obtaining multiple pooled ticket features of the recommended traffic ticket, the embodiments of the present application may perform difference processing on the ticket features of the candidate alternative search information and the pooled ticket features of the recommended traffic ticket in each feature dimension, so as to obtain the feature difference of the ticket features of the candidate alternative search information in each feature dimension, that is, the feature difference of each ticket feature of the candidate alternative search information.

[0141] In an alternative implementation, embodiments of the present application may perform a difference operation on the ticket features of the candidate alternative search information and the pooled ticket features of the recommended transportation tickets. For example, for any feature dimension, subtract the ticket features of the candidate alternative search information from the pooled ticket features of the recommended transportation tickets to obtain the feature difference of the ticket features of the candidate alternative search information in this feature dimension. Furthermore, by performing the difference operation on the ticket features of the candidate alternative search information and the pooled ticket features of the recommended transportation tickets in each feature dimension, the feature differences of each ticket feature of the candidate alternative search information can be obtained. For example, in the price dimension, perform a difference operation on the price feature of the candidate alternative search information and the pooled price feature of the recommended transportation tickets to obtain the feature difference of the price feature of the candidate alternative search information, simply referred to as the price feature difference; in the distance dimension, perform a difference operation on the distance feature of the candidate alternative search information and the pooled distance feature of the recommended transportation tickets to obtain the feature difference of the distance feature of the candidate alternative search information, simply referred to as the distance feature difference; in the time dimension, perform a difference operation on the time feature of the candidate alternative search information and the pooled time feature of the recommended transportation tickets to obtain the feature difference of the time feature of the candidate alternative search information, simply referred to as the time feature difference.

[0142] In an alternative implementation, as Figure 7 shown, the monotonic balance network module 430 may include: a plurality of difference calculation units 730, Figure 7 Taking 3 difference calculation units as an example, the plurality of difference calculation units 730 are used to implement step S620, that is, the plurality of difference calculation units 730 can perform a difference operation on each ticket feature of the candidate alternative search information and each pooled ticket feature of the recommended transportation tickets respectively to obtain the feature differences of each ticket feature of the candidate alternative search information. In an alternative implementation, one difference calculation unit can perform a difference operation on the ticket features of the candidate alternative search information and the pooled ticket features of the recommended transportation tickets in one feature dimension to obtain the feature difference of the ticket features of the candidate alternative search information in one feature dimension. Thus, the plurality of difference calculation units can perform a difference operation on the ticket features of the candidate alternative search information and the pooled ticket features of the recommended transportation tickets in multiple feature dimensions respectively to obtain the feature differences of the ticket features of the candidate alternative search information in multiple feature dimensions respectively, that is, the feature differences of each ticket feature of the candidate alternative search information.

[0143] For example, taking the multiple ticket features including price feature, distance feature, and time feature as an example, the plurality of difference calculation units may include:

[0144] A difference calculation unit that performs a difference operation on the price feature of the candidate alternative search information and the pooled price feature of the recommended transportation tickets, and this difference calculation unit outputs the price feature difference of the candidate alternative search information;

[0145] A difference calculation unit that calculates the difference between the distance feature of the candidate alternative search information and the pooled distance feature of the recommended transportation ticket, and the difference calculation unit outputs the distance feature difference of the candidate alternative search information;

[0146] A difference calculation unit that calculates the difference between the time feature of the candidate alternative search information and the pooled time feature of the recommended transportation ticket, and the difference calculation unit outputs the time feature difference of the candidate alternative search information;

[0147] For ease of explanation, for any candidate alternative search information, in the embodiment of the present application, the price feature difference of the candidate alternative search information can be defined as Δp, the distance feature difference as Δd, and the time feature difference as Δt.

[0148] Step S630: Quantize the feature differences of each ticket feature of the candidate alternative search information into binary vectors respectively, so as to obtain the binary vectors corresponding to the feature differences of each ticket feature of the candidate alternative search information.

[0149] After obtaining the feature differences of each ticket feature of the candidate alternative search information (such as the price feature difference Δp, the distance feature difference Δd, and the time feature difference Δt) in the embodiment of the present application, the feature differences of each ticket feature can be quantized respectively, so as to quantize the feature differences of each ticket feature into corresponding binary vectors, and obtain the binary vectors corresponding to the feature differences of each ticket feature of the alternative search information.

[0150] In an optional implementation, the feature differences of each ticket feature can be quantized through binning and reverse operations to obtain the corresponding binary vectors. Specifically, for any ticket feature, in the embodiment of the present application, the bin of the feature difference of the ticket feature of the candidate alternative search information can be determined from multiple pre-divided bins of the feature difference of the ticket feature; and based on the original binary vectors corresponding to each bin of the feature difference, the original binary vector corresponding to the bin where the feature difference is located is determined to obtain the original binary vector corresponding to the feature difference of the ticket feature of the candidate alternative search information; furthermore, the original binary vector corresponding to the feature difference of the ticket feature of the candidate alternative search information is reversed to obtain the binary vector corresponding to the feature difference of the ticket feature of the candidate alternative search information.

[0151] Exemplarily, taking the quantization process of the price feature difference Δp of the candidate alternative search information as an example, in the embodiments of the present application, multiple price feature difference levels can be pre-divided. Taking the pre-division of P price feature difference levels as an example, the range of the price feature difference is divided into P intervals, such as from the first price feature difference level to the P-th price feature difference level, corresponding to the P intervals into which the price feature difference is divided; Exemplarily, assuming that P is 3, 3 price feature difference levels can be pre-divided, that is, the range of the price feature difference is divided into 3 intervals. For example, the first price feature difference level is the interval where the price difference is between 0 and 20 yuan, the second price feature difference level is the interval where the price difference is between 21 yuan and 50 yuan, and the third price feature difference level is the interval where the price difference is between 51 yuan and 100 yuan; Of course, the above specific values and level divisions are only for illustrative purposes, and the number of specific price feature difference levels and the price difference intervals corresponding to each price feature difference level can be set according to actual situations, and the embodiments of the present application do not limit them;

[0152] Furthermore, for any price feature difference level, in the embodiments of the present application, a price feature difference level index corresponding to the price feature difference level can be set to identify each of the pre-divided price feature difference levels. For example, the price feature difference level indexes of the first price feature difference level to the P-th price feature difference level can be sequentially identified as price feature difference level index 1 to P, that is, P price feature difference levels correspond to P price feature difference level indexes;

[0153] In the embodiments of the present application, the original binary vectors corresponding to each price feature difference level can be further set, that is, each price feature difference level corresponds to an original binary vector respectively; thereby, in the embodiments of the present application, the price feature difference level in which the price feature difference Δp of the candidate alternative search information is located can be determined from the multiple pre-divided price feature difference levels, and based on the original binary vectors corresponding to each price feature difference level, the original binary vector corresponding to the price feature difference level in which the price feature difference of the candidate alternative search information is located can be determined to obtain the original binary vector corresponding to the price feature difference of the candidate alternative search information; For ease of explanation, the original binary vector corresponding to the price feature difference of the candidate alternative search information can be expressed as ;

[0154] Exemplarily, assuming that the price feature difference Δp of the candidate alternative search information is in the first price feature difference level, and the corresponding price feature difference level index is 1, then the original binary vector corresponding to the first price feature difference level is the original binary vector corresponding to the price feature difference of the candidate alternative search information ; Assuming that the original binary vector corresponding to the first price feature difference level is [0, 0,... 0, 1,..., 1], where the first terms are all 0 and the subsequent terms are 1, then is [0, 0, ..., 0, 1, ..., 1];

[0155] After obtaining the original binary vector corresponding to the price feature difference of the candidate alternative search information the embodiment of the present application can reverse the original binary vector For example, reverse the 1s in the original binary vector to 0s and the 0s to 1s in the original binary vector to obtain the binary vector corresponding to the price feature difference of the candidate alternative search information; for ease of explanation, the binary vector corresponding to the price feature difference of the candidate alternative search information can be expressed as i.e., is obtained by reversing

[0156] It can be understood that after reversing if the number of 1s in the binary vector corresponding to the price feature difference of a certain candidate alternative search information is more, compared with other candidate alternative search information, the price difference between the price feature of this candidate alternative search information and the pooled price feature (such as the minimum price feature) of the recommended transportation ticket is larger, and the price feature of this candidate alternative search information is cheaper, which is more in line with the user's price demand. Therefore, by quantifying the price feature difference of the candidate alternative search information into the binary vector the influence of the price feature difference on the finally recommended alternative search information can be considered more accurately when making alternative transportation ticket recommendations.

[0157] The quantization processing of the distance feature difference Δd and the time feature difference Δt of the candidate alternative search information is the same as above, and the corresponding simplified description is as follows.

[0158] In an alternative implementation, the embodiment of the present application can pre-divide multiple distance feature difference grades. For example, pre-divide D distance feature difference grades. At the same time, each distance feature difference grade corresponds to a distance feature difference grade index, and each distance feature difference grade is set with a corresponding original binary vector; thus, the embodiment of the present application can determine the distance feature difference grade in which the distance feature difference Δd of the candidate alternative search information is located from the pre-divided multiple distance feature difference grades, and based on the original binary vectors corresponding to each distance feature difference grade, determine the original binary vector corresponding to the distance feature difference grade in which the distance feature difference of the candidate alternative search information is located, so as to obtain the original binary vector corresponding to the distance feature difference of the candidate alternative search information; for ease of explanation, the original binary vector corresponding to the distance feature difference of the candidate alternative search information can be expressed as ; furthermore, the embodiment of the present application can reverse For example, reverse ​Reverse 1 to 0 and 0 to 1 in it to obtain the binary vector corresponding to the distance feature difference of the candidate alternative search information. For ease of explanation, the binary vector corresponding to the distance feature difference of the candidate alternative search information can be expressed as , that is Obtained by performing a reverse operation on . It should be noted that, for the binary vector corresponding to the distance feature difference of a certain candidate alternative search information , the more 0s in it, compared with other candidate alternative search information, the smaller the distance difference between this candidate alternative search information and the recommended traffic ticket, that is, the closer the distance of this candidate alternative search information.

[0159] In an alternative implementation, embodiments of the present application can pre-divide multiple time feature difference levels. For example, pre-divide T time feature difference levels. At the same time, each time feature difference level corresponds to a time feature difference level index, and each time feature difference level is set with a corresponding original binary vector. Thus, embodiments of the present application can determine the time feature difference level in which the time feature difference Δt of the candidate alternative search information is located from the pre-divided multiple time feature difference levels, and based on the original binary vectors corresponding to each time feature difference level, determine the original binary vector corresponding to the time feature difference level in which the time feature difference of the candidate alternative search information is located, so as to obtain the original binary vector corresponding to the time feature difference of the candidate alternative search information. For ease of explanation, the original binary vector corresponding to the time feature difference of the candidate alternative search information can be expressed as ; furthermore, embodiments of the present application can perform a reverse operation on , for example, reverse 1 to 0 and 0 to 1 in to obtain the binary vector corresponding to the time feature difference of the candidate alternative search information. For ease of explanation, the binary vector corresponding to the time feature difference of the candidate alternative search information can be expressed as , that is Obtained by performing a reverse operation on . It should be noted that, for the binary vector corresponding to the time feature difference of a certain candidate alternative search information , the more 0s in it, compared with other candidate alternative search information, the smaller the time difference between this candidate alternative search information and the recommended traffic ticket, that is, the closer the time of this candidate alternative search information.

[0160] In an alternative implementation, as shown in Figure 7 , the monotonic balance network module 430 may include: multiple monotonic encoders 740, Figure 7Taking three monotonic encoders as an example, multiple monotonic encoders 740 are used to implement step S630. That is, multiple monotonic encoders can respectively quantify the feature differences of each ticket feature of the candidate alternative search information into binary vectors, so as to obtain the binary vectors corresponding to the feature differences of the candidate alternative search information in each ticket feature. Specifically, one monotonic encoder can quantify the feature difference of one ticket feature of the candidate alternative search information into a corresponding binary vector, so that multiple monotonic encoders can output the binary vectors corresponding to the feature differences of the candidate alternative search information in multiple ticket features respectively.

[0161] In an alternative implementation example, taking multiple ticket features including price feature, distance feature, and time feature as an example, the multiple monotonic encoders may include:

[0162] A monotonic encoder that quantifies the price feature difference Δp of the candidate alternative search information into a binary vector corresponding to the price feature difference of the candidate alternative search information ;

[0163] A monotonic encoder that quantifies the distance feature difference Δd of the candidate alternative search information into a binary vector corresponding to the distance feature difference of the candidate alternative search information ;

[0164] A monotonic encoder that quantifies the time feature difference Δt of the candidate alternative search information into a binary vector corresponding to the time feature difference of the candidate alternative search information ;

[0165] Step S640: Determine the scores of the feature differences of the candidate alternative search information corresponding to each ticket feature according to the binary vectors corresponding to the feature differences of the candidate alternative search information in each ticket feature.

[0166] After obtaining the binary vectors corresponding to the feature differences of the candidate alternative search information in each ticket feature, the embodiments of the present application can determine the scores of the feature differences of the candidate alternative search information corresponding to each ticket feature based on the binary vectors corresponding to the feature differences of the candidate alternative search information in each ticket feature.

[0167] In an alternative implementation, for any candidate alternative search information, the embodiments of the present application can combine Figure 3Determine the scores of the feature differences of the candidate alternative search information corresponding to each ticket feature based on the search condition vector combination obtained in step S340 and the search result vector combination obtained in step S350. For example, combine the search condition vector combination and the search result vector combination with the binary vectors corresponding to the feature differences of the candidate alternative search information for each ticket feature, respectively, so as to obtain the scores of the feature differences of the candidate alternative search information corresponding to each ticket feature. By way of example, for any candidate alternative search information, in an embodiment of the present application, the search condition vector combination and the search result vector combination may be respectively subjected to a dot product operation with the binary vectors corresponding to the feature differences of the candidate alternative search information for each ticket feature to obtain the scores of the feature differences of the candidate alternative search information corresponding to each ticket feature.

[0168] As an implementation example, for any candidate alternative search information, assume that the search condition vector combination is and the search result vector combination is . Then, taking the scores of the price feature difference Δp, the distance feature difference Δd, and the time feature difference Δt of the candidate alternative search information as examples, when determining the score of the price feature difference Δp, , and (the binary vector corresponding to the price feature difference of the candidate alternative search information) may be subjected to a dot product operation to obtain the score of the price feature difference Δp. Assume that the score of the candidate alternative search information for the price feature difference Δp is Logit p . Then Logit p can be expressed as: ;

[0169] When determining the score of the distance feature difference Δd, , and (the binary vector corresponding to the distance feature difference of the candidate alternative search information) may be subjected to a dot product operation to obtain the score of the distance feature difference Δd. Assume that the score of the candidate alternative search information for the distance feature difference Δd is Logit d . Then Logit d can be expressed as: ;

[0170] When determining the score of the time feature difference Δt, , and (the binary vector corresponding to the time feature difference of the candidate alternative search information) may be subjected to a dot product operation to obtain the score of the time feature difference Δt. Assume that the score of the candidate alternative search information for the time feature difference Δt is Logit t . Then Logit t can be expressed as: 。

[0171] In an alternative implementation, in combination with Figure 7 as shown, the monotonic balance network module 430 may include: a plurality of feature difference score units 750, Figure 7 Taking 3 feature difference score units as an example, the plurality of feature difference score units 750 are used to implement step S640. Specifically, one feature difference score unit can determine the score of the candidate alternative search information in the feature difference corresponding to one ticket feature. For example, one feature difference score unit can determine the score of the candidate alternative search information in the feature difference corresponding to one ticket feature according to the binary vector corresponding to the feature difference of the candidate alternative search information in one ticket feature, so that the plurality of feature difference score units can output the scores of the candidate alternative search information in the feature differences corresponding to multiple ticket features, that is, the scores of the candidate alternative search information in the feature differences corresponding to each ticket feature.

[0172] In an alternative implementation, the feature difference score unit may be a dot product unit, which is used to perform a dot product operation on the search condition vector combination and the search result vector combination, and the binary vector corresponding to the feature difference of the candidate alternative search information in the ticket feature. Thus, by arranging a plurality of dot product units, the search condition vector combination and the search result vector combination can be respectively subjected to dot product operations with the binary vectors corresponding to the feature differences of the candidate alternative search information in each ticket feature to obtain the scores of the candidate alternative search information in the feature differences corresponding to each ticket feature.

[0173] Exemplarily, taking the multiple ticket features including price feature, distance feature, and time feature as an example, the plurality of dot product units may include:

[0174] A dot product unit that performs a dot product operation on the search condition vector combination , the search result vector combination , and the binary vector corresponding to the price feature difference of the candidate alternative search information . This dot product unit outputs the score Logit of the candidate alternative search information in the price feature difference p ;

[0175] A dot product unit that performs a dot product operation on the search condition vector combination , the search result vector combination , and the binary vector corresponding to the distance feature difference of the candidate alternative search information . This dot product unit outputs the score Logit of the candidate alternative search information in the distance feature difference d ;

[0176] A dot product unit that performs a dot product operation on the search condition vector combination , the search result vector combination , the binary vector corresponding to the time feature difference of the candidate alternative search information A dot product unit that performs a dot product operation, and the dot product unit outputs the score Logit of the candidate alternative search information in the time feature difference t .

[0177] Step S650: Synthesize the scores of the candidate alternative search information in the feature differences corresponding to each ticket feature to obtain the feature difference score of the candidate alternative search information

[0178] After obtaining the scores of the candidate alternative search information in the feature differences corresponding to each ticket feature, for example, obtaining the score Logit of the candidate alternative search information in the price feature difference p , the score Logit in the distance feature difference d , the score Logit in the time feature difference t After that, the embodiments of the present application can synthesize the scores of the candidate alternative search information in the feature differences corresponding to each ticket feature, such as weighted summation, to obtain the feature difference score of the candidate alternative search information

[0179] In an alternative implementation, the embodiments of the present application can preset the weights of each ticket feature, and then perform weighted summation processing based on the weights of each ticket feature and the scores of the candidate alternative search information in the feature differences corresponding to each ticket feature to obtain the feature difference score of the candidate alternative search information

[0180] Exemplarily, taking multiple ticket features including price feature, distance feature, and time feature as an example, the embodiments of the present application can preset the weight of the price feature (denoted as w p ), the weight of the distance feature (denoted as w d ), and the weight of the time feature (denoted as w t ), so that the weight w of the price feature p is multiplied by the score Logit of the candidate alternative search information in the price feature difference p , the weight w of the distance feature d is multiplied by the score Logit of the candidate alternative search information in the distance feature difference d , the weight w of the time feature t is multiplied by the score Logit of the candidate alternative search information in the time feature difference t , and then the products of each item are added to complete the weighted summation processing to obtain the feature difference score of the candidate alternative search information

[0181] Exemplarily, the feature difference score of the candidate alternative search information can be regarded as a score output by the alternative transportation ticket recommendation subsystem of the travel service platform using the model, denoted as Logit 2 , and .

[0182] In an alternative implementation, in combination with Figure 7 As shown, the monotonic balance network module 430 may include: a feature difference score integration unit 760, and the feature difference score integration unit 760 is used to implement step S650, that is, the feature difference score integration unit 760 may integrate the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information to obtain the feature difference score of the candidate alternative search information.

[0183] In an alternative implementation, the feature difference score integration unit 760 may be a weighted summation unit, which is used to perform weighted summation processing based on the weights of each ticket feature and the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information to obtain the feature difference score of the candidate alternative search information. For example, the weighted summation unit may be based on the weight w of the price feature p and the score Logit of the candidate alternative search information in the price feature difference p , the weight w of the distance feature d and the score Logit of the candidate alternative search information in the distance feature difference d , and the weight w of the time feature t and the score Logit of the candidate alternative search information in the time feature difference t , perform weighted summation processing, and output the feature difference score of the candidate alternative search information.

[0184] Each candidate alternative search information passes through the Figure 6 process shown, and then the feature difference score Logit of each candidate alternative search information can be obtained 2 . The higher the Logit of the candidate alternative search information 2 , the closer the candidate alternative search information is to the user's needs.

[0185] Returning to Figure 2 As shown, in step S260, according to the search preference scores of the user for each candidate alternative search information and the feature difference scores of each candidate alternative search information, the scores of each candidate alternative search information are determined.

[0186] For any candidate alternative search information, after determining the search preference score Logit of the user for the candidate alternative search information 1 , and the feature difference score Logit of the candidate alternative search information 2 , the embodiments of the present application may at least integrate the search preference score Logit of the candidate alternative search information 1 , and the feature difference score Logit of the candidate alternative search information 2 , to obtain the score of the candidate alternative search information.

[0187] In an alternative implementation, when determining the score of candidate alternative search information, embodiments of the present application may further combine a search preference weight and a set bias term parameter. The search preference weight is the weight corresponding to the search preference score and can be represented as w 1 . The bias term parameter is a bias term introduced to prevent user accidental touch or misoperation and can be represented as . Specifically, the bias term parameter is an additional constant term used to adjust the model output. In a recommendation system, the bias term parameter can be a constant that prevents misoperation or excessive bias towards certain features, and it ensures the rationality of the output by adjusting the score, avoiding unreasonable results caused by user accidental touch or excessive feature bias.

[0188] In an alternative implementation, for any candidate alternative search information, embodiments of the present application may multiply the search preference weight w 1 by the search preference score Logit of the user for the candidate alternative search information 1 , add the product to the feature difference score Logit of the candidate alternative search information 2 and the set bias term parameter, so as to obtain the score of the candidate alternative search information.

[0189] For example, if the score of a certain candidate alternative search information is Logit, then Logit is expressed as:

[0190] .

[0191] In a further alternative implementation, the sum of the search preference weight and the weights of each ticket feature is 1; for example, the sum of the search preference weight w 1 , the weight w p of the price feature, the weight w d of the distance feature, and the weight w t of the time feature can be 1, that is, w 1 +w p +w d +w t =1, that is to say, the combined influence of the above four weights is equal to 1.

[0192] In a further alternative implementation, embodiments of the present application may adjust the values of each weight according to the actual situation and data characteristics. For example, when the search preference is more important and the influence of the search preference weight w 1 on the recommendation is greater than other weights, the search preference weight w 1 can be set to be the largest. For example, w 1 =0.5, w p =0.2, w d =0.2, w t= 0.1; for another example, when the price feature is more important, such as for price-sensitive users, the weight w of the price feature p has a greater impact on the recommendation than other weights, then the weight w of the price feature can be set p to be the largest, such as w 1 = 0.2, w p = 0.5, w d = 0.2, w t = 0.1; in some scenarios, the alternative transportation ticket recommendation of the travel service platform may focus on distance and time, then the weight w of the distance feature d , the weight w of the time feature t may have relatively large weight values; of course, it is also possible to consider the above four weights to be equally important, so as to set balanced weights, such as the above four weights being equal, all 0.25.

[0193] In an alternative implementation, as shown in Figure 4 , the alternative transportation ticket recommendation subsystem 400 may include: a final score integration module 440, and the final score integration module 440 may be used to implement step S260, that is, the final score integration module may determine the scores of each candidate alternative search information according to the search preference scores of the user for each candidate alternative search information and the feature difference scores of each candidate alternative search information.

[0194] Exemplarily, the final score integration module may be in the form of a weighted summation unit, and specifically may be used for any candidate alternative search information, multiplying the search preference weight w 1 by the search preference score Logit of the user for the candidate alternative search information 1 , adding the product to the feature difference score Logit of the candidate alternative search information 2 and the set bias term parameter, so as to obtain the score Logit of the candidate alternative search information.

[0195] Returning to Figure 2 shown in, in step S270, according to the scores of each candidate alternative search information, the recommended alternative search information is determined from multiple candidate alternative search information.

[0196] After obtaining the scores Logit of each candidate alternative search information, the embodiments of the present application may, based on the scores of each candidate alternative search information, determine the alternative search information whose scores meet the recommendation requirements from multiple candidate alternative search information as the recommended alternative search information.

[0197] In an alternative implementation, embodiments of the present application may sort the multiple candidate alternative search information according to their scores. The sorting may be in descending order (i.e., the candidate alternative search information with the highest score is ranked first) or in ascending order. Then, according to the sorting result, a set number of candidate alternative search information is selected from the candidate alternative search information with the highest score as the recommended alternative search information. The number of the recommended alternative search information (i.e., the set number of recommendations) can be set according to specific circumstances, and embodiments of the present application do not limit it.

[0198] In other alternative implementations, embodiments of the present application may also support using the candidate alternative search information whose scores meet the score criteria as the recommended alternative search information. For example, the candidate alternative search information with a score higher than the score criteria is used as the recommended alternative search information.

[0199] Furthermore, the recommended alternative search information and the search results of the recommended transportation tickets can be displayed together on the search result page. In a further alternative implementation, the display method of the recommended alternative search information on the search result page includes, but is not limited to, any one of horizontal sliding display, list display, card display, etc. Among them, horizontal sliding display means that the user can view the recommended alternative search information on the search result page by dragging, scrolling, or clicking an arrow to slide horizontally on the screen. For example, when the recommended alternative search information is displayed by horizontal sliding, if there are multiple recommended alternative search information, the multiple recommended alternative search information will not be displayed all at once on the search result page, but presented in a horizontal sliding manner, that is, the user can view all the recommended alternative search information by sliding. Horizontal sliding display can be applied to the search result pages of mobile devices or web pages to display the recommended alternative search information.

[0200] In an alternative implementation, as shown in Figure 4 the alternative transportation ticket recommendation subsystem 400 may include: a final recommendation module 450. The final recommendation module 450 is used to implement step S270, that is, the final recommendation module can determine the recommended alternative search information from multiple candidate alternative search information according to the scores of each candidate alternative search information.

[0201] When the ticket recommendation method provided by the embodiment of the present application is used for a user to conduct a ticket search, in addition to recommending search results of recommended transportation tickets that meet the ticket search conditions, it also recommends alternative search information of alternative transportation tickets related to the ticket search conditions, so as to facilitate the user to search for alternative transportation tickets. When recommending the alternative search information, the embodiment of the present application can first determine multiple candidate alternative search information of alternative transportation tickets related to the ticket search conditions. Each candidate alternative search information includes a ticket search condition for searching alternative transportation tickets and is associated with multiple ticket characteristics related to the user's ticket decision-making. Then, score each candidate alternative search information. In the embodiment of the present application, the score of the candidate alternative search information includes the search preference score of the user for the candidate alternative search information and the feature difference score of the candidate alternative search information. The search preference score of the user for the candidate alternative search information can reflect the preference degree of the user for selecting the candidate alternative search information when conducting a ticket search. The feature difference score of the candidate alternative search information takes into account the difference between the candidate alternative search information and the recommended transportation tickets in terms of ticket characteristics related to the user's ticket decision-making. Through the difference, the deviation degree of the alternative transportation ticket corresponding to the candidate alternative search information from the recommended transportation ticket in multiple dimensions of ticket characteristics can be effectively measured, thereby helping the travel service platform screen and recommend the alternative transportation ticket closest to the user's needs and improving the recommendation accuracy of the alternative transportation ticket recommendation.

[0202] Therefore, the embodiment of the present application can determine the search preference score of the user for the candidate alternative search information according to the user's user characteristic information, historical behavior information, the ticket search condition of the candidate alternative search information, and the search result of the recommended transportation ticket; determine the score of the feature difference of the candidate alternative search information compared with the recommended transportation ticket for each ticket characteristic based on multiple ticket characteristics related to the user's ticket decision-making, and obtain the feature difference score of the candidate alternative search information based on the score of the feature difference of the candidate alternative search information for each ticket characteristic; further, determine the score of each candidate alternative search information according to the search preference score of the user for each candidate alternative search information and the feature difference score of each candidate alternative search information; finally, determine the recommended alternative search information from multiple candidate alternative search information according to the score of each candidate alternative search information.

[0203] It can be seen that the embodiment of the present application combines the search preference score of the user for the candidate alternative search information and the feature difference score of the candidate alternative search information, which can comprehensively consider the user's search preference and the actual situation of the alternative transportation ticket corresponding to the candidate alternative search information, generate an accurate score for the candidate alternative search information. Furthermore, this dual evaluation system of the search preference score and the feature difference score of the candidate alternative search information can reduce the error caused by one-sided consideration of a single factor, more accurately recommend alternative search information for alternative transportation tickets for users, and improve the recommendation accuracy of alternative transportation ticket recommendations; thus providing users with more choice possibilities for travel, enhancing the travel experience of users, and further increasing the ticket click-through rate and conversion rate of the travel service platform.

[0204] Furthermore, in the construction of the alternative transportation ticket recommendation subsystem, the embodiment of the present application introduces a monotonic balance network architecture, that is, a monotonic balance network module, so as to model the recommendation influence of multiple ticket features (such as price feature, distance feature, time feature) related to the user's ticket decision. Furthermore, the model can be smoothly adjusted based on the differences in ticket features to enhance the accuracy of the recommendation.

[0205] Furthermore, the score of the candidate alternative search information determined by the embodiment of the present application is the Logit fusion result of multiple models, specifically the fusion result of the Logit score of the main network module and the Logit score of the monotonic balance network module. By obtaining the score of the candidate alternative search information through the Logit fusion of multiple models, the negative impact caused by overfitting of a certain model or being too sensitive to certain features can be reduced, thereby enhancing stability; that is to say, the Logit result of a single model may be affected by outliers or be unstable on certain features, while through the Logit fusion of multiple models, the Logit results of multiple models can be integrated through a weighted average or voting mechanism, thereby improving the accuracy and robustness of the recommendation.

[0206] Taking the recommendation of nearby tickets as an example, compared with the use of traditional deep neural network (DNN) algorithm for nearby ticket recommendation, the use of the ticket recommendation scheme provided in the embodiment of the present application for nearby ticket recommendation can have obvious effects on the increase of AUC (Area Under Curve) and CTR (Click Through Rate); specifically, through offline experiments, compared with the use of DNN algorithm for nearby ticket recommendation, the ticket recommendation scheme provided in the embodiment of the present application has a 3.4% AUC increase in nearby ticket recommendation, which improves the ability of the travel service platform to distinguish different user needs and preferences, thereby improving the recommendation accuracy of alternative transportation ticket recommendations; through online experiments, compared with the use of DNN algorithm for nearby ticket recommendation, the ticket recommendation scheme provided in the embodiment of the present application has a 14.1% CTR increase in nearby ticket recommendation, which improves the user's click-through rate.

[0207] In an optional implementation, the embodiment of the present application further provides a ticket recommendation system, which is applied to a travel service platform. Figure 8 An optional block diagram of the ticket recommendation system provided by the embodiment of the present application is shown exemplarily. Figure 8 The ticket recommendation system provided in the embodiment of the present application may include:

[0208] A request acquisition module 810, configured to acquire a user's ticket search request, wherein the ticket search request includes a ticket search condition;

[0209] A ticket search engine 820, for determining search results of recommended transportation tickets that meet the ticket search conditions;

[0210] An alternative transportation ticket recommendation subsystem 400 is used to determine multiple candidate alternative search information related to alternative transportation tickets for the ticket search conditions. Each candidate alternative search information includes the ticket search conditions for searching alternative transportation tickets and is associated with multiple ticket features related to the user's ticket decision-making. For any candidate alternative search information, based on the user's user feature information, historical behavior information, the ticket search conditions of the candidate alternative search information, and the search results of the recommended transportation tickets, determine the search preference score of the user for the candidate alternative search information. And, for any candidate alternative search information, based on the multiple ticket features, determine the score of the feature difference corresponding to each ticket feature of the candidate alternative search information compared to the recommended transportation ticket, and based on the scores of the feature differences corresponding to each ticket feature of the candidate alternative search information, obtain the feature difference score of the candidate alternative search information. According to the search preference scores of the user for each candidate alternative search information and the feature difference scores of each candidate alternative search information, determine the scores of each candidate alternative search information. According to the scores of each candidate alternative search information, determine the recommended alternative search information from the multiple candidate alternative search information.

[0211] In an alternative implementation, the relevant components and functions of the alternative transportation ticket recommendation subsystem 400 can be referred to the previous description and will not be elaborated here.

[0212] In a further alternative implementation, an embodiment of the present application also provides a travel server, which is regarded as a server or a server cluster set in the travel service platform. In an alternative implementation, the travel server may include a memory and a processor. The memory stores computer execution instructions, and the processor calls the computer execution instructions stored in the memory to execute the ticket recommendation method provided by the embodiment of the present application.

[0213] In a further alternative implementation, an embodiment of the present application also provides a storage medium that stores computer execution instructions. When the computer execution instructions are executed (for example, when the computer execution instructions are executed by the processor), the ticket recommendation method provided by the embodiment of the present application is implemented.

[0214] In a further alternative implementation, an embodiment of the present application also provides a computer program product, including computer execution instructions. When the computer execution instructions are executed (for example, when the computer execution instructions are executed by the processor), the ticket recommendation method provided by the embodiment of the present application is implemented.

[0215] The above describes multiple embodiment solutions provided by the embodiments of the present application. The optional ways introduced in each embodiment solution can be combined and cross-referenced with each other without conflict, so as to extend multiple possible embodiment solutions, which can all be regarded as the embodiment solutions disclosed and made public by the embodiments of the present application.

[0216] Although the embodiments of the present application are disclosed as above, the present application is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application shall be subject to the scope defined by the claims.

Claims

1. A ticket recommendation method, characterized in that: include: Obtaining a ticket search request from a user, wherein the ticket search request includes a ticket search condition; Determining search results of recommended transportation tickets that match the ticket search criteria; Determine a plurality of candidate replacement search information of alternative transportation tickets related to the ticket search condition, each candidate replacement search information includes a ticket search condition for searching for alternative transportation tickets and is associated with a plurality of ticket features related to the user's ticket decision; For any candidate alternative search information, determine the user's search preference score for the candidate alternative search information based on the user's user characteristic information, historical behavior information, ticket search conditions of the candidate alternative search information, and search results of recommended transportation tickets; And, for any candidate alternative search information, based on the multiple ticket features, determine the score of the feature difference between the candidate alternative search information and the recommended transportation ticket in each ticket feature, and based on the score of the feature difference between the candidate alternative search information and the recommended transportation ticket in each ticket feature, obtain the feature difference score of the candidate alternative search information; the step of determining the score of the feature difference between the candidate alternative search information and the recommended transportation ticket in each ticket feature based on the multiple ticket features includes: performing difference processing between the ticket features of the candidate alternative search information and the pooled ticket features of the recommended transportation ticket in each feature dimension to obtain the feature difference of each ticket feature of the candidate alternative search information; based on the binary vector corresponding to the feature difference between the candidate alternative search information and each ticket feature, determine the score of the feature difference between the candidate alternative search information and each ticket feature; Determining the score of each candidate alternative search information according to the user's search preference score for each candidate alternative search information and the feature difference score of each candidate alternative search information; According to the scores of each candidate alternative search information, recommended alternative search information is determined from the plurality of candidate alternative search information.

2. The method according to claim 1, characterized in that The difference processing of the ticket features of the candidate alternative search information and the pooled ticket features of the recommended transportation tickets in each feature dimension to obtain the feature differences of each ticket feature of the candidate alternative search information includes: Pooling multiple ticket features associated with each search result of recommended transportation tickets to obtain multiple pooled ticket features of recommended transportation tickets; Performing difference processing on each ticket feature of the candidate replacement search information and each pooled ticket feature of the recommended transportation ticket, respectively, to obtain the feature difference of each ticket feature of the candidate replacement search information; The determining of the score of the feature difference of the candidate alternative search information corresponding to each ticket feature based on the binary vector corresponding to the feature difference of the candidate alternative search information in each ticket feature includes: quantizing the feature differences of each ticket feature of the candidate alternative search information into binary vectors respectively, so as to obtain the binary vectors corresponding to the feature differences of each ticket feature of the candidate alternative search information; According to the binary vectors corresponding to the feature differences of the candidate replacement search information in each ticket feature, the scores of the feature differences of the candidate replacement search information corresponding to each ticket feature are determined respectively.

3. The method according to claim 2, characterized in that The feature differences of each ticket feature of the candidate alternative search information are quantized into binary vectors respectively to obtain binary vectors corresponding to the feature differences of each ticket feature of the candidate alternative search information, including: For any ticket feature, determine the feature difference level at which the feature difference of the ticket feature of the candidate replacement search information is located from a plurality of feature difference levels pre-divided by the ticket feature; Based on the original binary vectors corresponding to the various feature difference levels, determining the original binary vector corresponding to the feature difference level in which the candidate alternative search information is located, so as to obtain the original binary vector corresponding to the feature difference of the ticket feature; Inverting the original binary vector corresponding to the feature difference of the ticket feature of the candidate replacement search information to obtain the binary vector corresponding to the feature difference of the ticket feature of the candidate replacement search information; The feature difference scores of the candidate alternative search information are obtained based on the feature difference scores corresponding to the various ticket features of the candidate alternative search information, including: Based on the weights of the various ticket features and the feature difference scores of the candidate replacement search information corresponding to the various ticket features, a weighted summation process is performed to obtain the feature difference scores of the candidate replacement search information.

4. The method according to claim 3, characterized in that Determining the user's search preference score for the candidate alternative search information based on the user's user feature information, historical behavior information, ticket search conditions of the candidate alternative search information, and search results of recommended transportation tickets includes: Embed each user feature information of the user into corresponding user feature vectors, and concatenate each user feature vector into a user feature vector combination; For any piece of historical behavior information of the user, embed each historical behavior feature in the historical behavior information into corresponding historical behavior feature vectors, and concatenate each historical behavior feature vector into a historical behavior feature vector combination to obtain a historical behavior feature vector combination corresponding to the historical behavior information; Based on the combination of historical behavior feature vectors corresponding to each piece of historical behavior information, the potential relationship of historical behavior information is extracted through the self-attention mechanism and converted into the user's historical behavior representation; Embed each ticket search condition of the candidate replacement search information into corresponding search condition vectors, and concatenate each search condition vector into a corresponding search condition vector combination; Embed each search result of the recommended transportation ticketing service into a corresponding search result vector, and concatenate each search result vector into a corresponding search result vector combination; Based on the user's historical behavior representation and the combination of search condition vectors, a comprehensive representation vector of ticket search conditions that integrates the user's historical behavior and candidate alternative search information is determined through a self-attention mechanism; The user feature vector combination, the comprehensive representation vector, the search condition vector combination, and the search result vector combination are combined into a combined vector; the combined vector is subjected to multi-layer feature mapping and nonlinear processing to obtain the user's search preference score for the candidate alternative search information.

5. The method according to claim 4, characterized in that The step of determining the scores of the feature differences of the candidate alternative search information corresponding to each ticket feature according to the binary vectors corresponding to the feature differences of the candidate alternative search information in each ticket feature comprises: Perform a dot multiplication operation on the combination of search condition vectors and the combination of search result vectors with the binary vectors corresponding to the feature differences of candidate alternative search information in each ticket feature, respectively, to obtain the scores of the feature differences of candidate alternative search information in each ticket feature; Determining the score of each candidate alternative search information according to the user's search preference score for each candidate alternative search information and the feature difference score of each candidate alternative search information includes: For any candidate alternative search information, the search preference weight is multiplied by the user's search preference score for the candidate alternative search information, and the product is added to the feature difference score of the candidate alternative search information and the set bias parameter to obtain the score of the candidate alternative search information; The sum of the search preference weight and the weight of each ticket feature is 1.

6. The method according to any one of claims 1 to 5, characterized in that: The multiple ticket features associated with the candidate alternative search information and relevant to the user's ticket decision include: a price feature, indicating the price of the alternative transportation ticket corresponding to the candidate alternative search information; A distance feature, which indicates the distance between the departure place and the target departure place of the alternative transportation ticket corresponding to the candidate alternative search information, and the distance between the destination and the target destination of the alternative transportation ticket corresponding to the candidate alternative search information; wherein the departure place when the user searches for tickets is the target departure place, and the destination when the user searches for tickets is the target destination; The time feature indicates the departure time of the alternative transportation ticket corresponding to the candidate alternative search information.

7. A ticket recommendation system, characterized in that: Used to execute the ticket recommendation method according to any one of claims 1 to 6, the ticket recommendation system comprises: A request acquisition module, used to acquire a user's ticket search request, wherein the ticket search request includes a ticket search condition; A ticket search engine, used to determine search results of recommended transportation tickets that meet the ticket search criteria; Alternative transportation ticketing recommendation subsystem for: Determine multiple candidate alternative search information of alternative transportation tickets related to the ticket search condition, each candidate alternative search information includes a ticket search condition for searching for alternative transportation tickets and is associated with multiple ticket features related to the user's ticketing decision; for any candidate alternative search information, determine the user's search preference score for the candidate alternative search information based on the user's user feature information, historical behavior information, the ticket search condition of the candidate alternative search information, and the search results of the recommended transportation tickets; and, for any candidate alternative search information, determine the score of the feature difference corresponding to each ticket feature of the candidate alternative search information compared with the recommended transportation ticket based on the multiple ticket features, and obtain the feature difference score of the candidate alternative search information based on the feature difference score corresponding to each ticket feature of the candidate alternative search information; determine the score of each candidate alternative search information based on the user's search preference score for each candidate alternative search information and the feature difference score of each candidate alternative search information; and determine the recommended alternative search information from the multiple candidate alternative search information based on the score of each candidate alternative search information.

8. The system according to claim 7, characterized in that The alternative transportation ticket recommendation subsystem includes: A recall module, configured to determine a plurality of candidate alternative search information of alternative transportation tickets related to the ticket search condition; The main network module is used to determine the user's search preference score for any candidate alternative search information based on the user's user characteristic information, historical behavior information, ticket search conditions of the candidate alternative search information, and search results of recommended transportation tickets; a monotone balance network module, for determining, for any candidate alternative search information, a score of a feature difference between the candidate alternative search information and the recommended transportation ticket corresponding to each ticket feature based on the plurality of ticket features, and obtaining a feature difference score of the candidate alternative search information based on the score of the feature difference between the candidate alternative search information and each ticket feature; A final score synthesis module, used to determine the score of each candidate alternative search information according to the user's search preference score for each candidate alternative search information and the feature difference score of each candidate alternative search information; The final recommendation module is used to determine the recommended alternative search information from multiple candidate alternative search information according to the scores of each candidate alternative search information.

9. A travel server, characterized in that: The system comprises a memory and a processor, wherein the memory stores computer-executable instructions, and the processor calls the computer-executable instructions to execute the ticket recommendation method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer-executable instructions, which, when executed, implement the ticket recommendation method according to any one of claims 1 to 6.

Images