Hotel room intelligent recommendation method based on artificial intelligence

Abstract

The application discloses a hotel room intelligent recommendation method based on artificial intelligence and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring in real time check-in request information sent by a user terminal, extracting a user behavior space-time sequence containing a time stamp and a spatial coordinate from the check-in request information; extracting a user preference feature vector from the user behavior space-time sequence, and constructing a multi-dimensional attribute space of user demand; acquiring real-time state data of a room provided by a hotel end, including spatial layout parameters, facility configuration parameters and environmental perception parameters; coupling and analyzing the multi-dimensional attribute space and the real-time state data of the room, and generating a matching degree dynamic distribution map; identifying a candidate room set meeting user demand; calculating a comprehensive recommendation index for each room in the candidate room set, wherein the index is based on the matching degree dynamic distribution map and a hotel operation strategy weight; and generating a recommendation list in the order of the comprehensive recommendation index and sending the recommendation list to the user terminal.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an intelligent hotel room recommendation method based on artificial intelligence. Background Technology

[0002] Hotel room recommendation methods are widely used in online travel and hotel management systems. Existing solutions primarily construct user profiles by collecting static attribute preferences from users' historical check-in records, such as room type, price range, or facility tags, and then match these profiles with the basic attributes of hotel rooms to generate a recommendation list. Some solutions incorporate user location information or browsing behavior to enhance the spatiotemporal relevance of recommendations. A drawback of existing technologies is that the spatiotemporal patterns in users' historical check-in data are not fully explored. Traditional preference extraction methods only count the frequency of attribute tags, ignoring the time intervals, order, and geographic clustering and migration patterns between check-in records. This results in insufficient characterization of user travel cycles and spatial preferences, making it difficult to predict users' deeper needs within specific time windows and geographic contexts. Furthermore, the utilization of room-side data often remains at the static level of room type information, failing to reflect real-time changes in room availability, specific environmental parameters, and facility configurations. This makes it easy for recommendations to include rooms that are already occupied, under maintenance, or whose environmental conditions do not meet user comfort requirements. Furthermore, the recommendation ranking is mainly based on matching scores or historical ratings, failing to incorporate hotel operational strategies such as price discounts and inventory pressures during specific periods. This leads to a disconnect between the recommended list and the hotel's actual operational goals. The problems that need to be addressed are: how to extract temporally periodic and spatially clustered behavioral sequences from users' discrete check-in history, and how to dynamically couple the resulting multi-dimensional demand space with real-time multi-dimensional room status data to obtain a spatiotemporal matching degree distribution; and how to integrate dynamically changing hotel operational strategy weights into the recommendation index calculation so that the ranking results balance user personalized matching with hotel revenue management needs. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent hotel room recommendation method based on artificial intelligence. This method can extract spatiotemporal behavior sequences from user check-in requests and construct a multi-dimensional attribute space. This space is then coupled with real-time multi-dimensional status data of the hotel rooms to form a dynamic distribution of matching degree. Furthermore, hotel operation strategy weights are incorporated into the comprehensive recommendation index to enhance the overall responsiveness of the recommendation to user spatiotemporal needs, real-time status of the hotel rooms, and hotel business objectives.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The purpose of the present invention is to provide an intelligent hotel room recommendation method based on artificial intelligence, so as to achieve high-precision dynamic matching between hotel rooms and user needs, thereby improving recommendation accuracy and user experience.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A hotel room intelligent recommendation method based on artificial intelligence, which acquires check-in request information sent by user terminals in real time, extracts a spatiotemporal sequence of user behavior from the check-in request information, the spatiotemporal sequence of user behavior including timestamps and spatial coordinates of the user's historical check-in records. Further, user preference feature vectors are extracted from the spatiotemporal sequence of user behavior, and a multi-dimensional attribute space of user needs is constructed accordingly. Simultaneously, real-time room status data provided by the hotel is acquired, including room spatial layout parameters, facility configuration parameters, and environmental perception parameters. The multi-dimensional attribute space is coupled and analyzed with the real-time room status data to generate a dynamic distribution map of the matching degree between rooms and users. Based on this dynamic distribution map, a set of candidate rooms that meet user needs is identified. A comprehensive recommendation index is calculated for each room in the candidate room set, the comprehensive recommendation index being obtained based on the dynamic distribution map of the matching degree and the hotel operation strategy weights. The rooms are sorted in descending order of the comprehensive recommendation index, a recommendation list is generated, and sent to the user terminal.

[0006] As a preferred technical solution of the present invention, the process of extracting the spatiotemporal sequence of user behavior from check-in request information includes: parsing the user identifier in the check-in request information and calling the user's historical check-in database; arranging each check-in record in the historical check-in database in chronological order and extracting the check-in timestamp and hotel geographical coordinates corresponding to each record; performing differential processing on the time intervals of adjacent check-in records to obtain a time interval sequence; performing periodic detection on the time interval sequence to identify the periodic pattern of user travel, and combining it with the spatial clustering of hotel geographical coordinates to form the spatiotemporal sequence of user behavior. By mining the spatiotemporal sequence, user travel patterns can be accurately captured.

[0007] As a further preferred technical solution of the present invention, when extracting user preference feature vectors from user behavior spatiotemporal sequences, hotels with user check-in times exceeding a preset threshold are first selected from the user behavior spatiotemporal sequences, and attribute tag sets for these hotels are obtained; the frequency of each attribute tag in the user's check-in history is counted, and the frequency is normalized and used as the weight value of the attribute tag; the attribute tag set and its corresponding weight value are combined into an initial feature vector; sparsity correction is performed on the initial feature vector, and duplicate or redundant attribute tags are deleted to obtain the user preference feature vector. This processing method effectively removes noisy tags, making preference characterization more accurate.

[0008] In constructing a multi-dimensional attribute space for user needs, the technical solution of this invention is as follows: User preference feature vectors are mapped to a preset hotel attribute classification system, which includes location, price, room type, service, and facility dimensions. Under each dimension, a demand intensity curve is generated based on the weight values ​​in the user preference feature vector. The demand intensity curves of each dimension are normalized and aligned to form a multi-dimensional attribute space under a unified coordinate system. Blank areas in the multi-dimensional attribute space are filled using interpolation to ensure that the demand intensity curves in each dimension are continuous and differentiable. The resulting attribute space comprehensively reflects the user's overall needs and their changing trends across multiple dimensions.

[0009] When acquiring real-time guest room status data, this invention collects the current occupancy, cleanliness, and maintenance status of each guest room in real time through the hotel management system interface; extracts spatial layout parameters for each guest room from the hotel space database, including room orientation, floor height, and area; extracts facility configuration parameters for each guest room from the hotel facility database, including bed type, bathroom fixtures, and a list of smart home devices; and collects environmental perception parameters for each guest room from an environmental sensor network, including temperature, humidity, and noise levels. This multi-dimensional real-time data provides a dynamic and comprehensive decision-making basis for subsequent matching.

[0010] In the coupling analysis stage, the preferred implementation method is as follows: The demand intensity curve for each dimension in the multi-dimensional attribute space is matched point-by-point with the actual value of the corresponding dimension in the real-time status data of guest rooms to obtain the local matching degree for each dimension; based on preset dimension weight coefficients, the local matching degrees of each dimension are weighted and summed to obtain the global matching degree; a spatial distance decay function is constructed with the target geographical location in the user's check-in request as the center, and the global matching degree is multiplied by the spatial distance decay coefficient to obtain the spatially corrected matching degree; this spatially corrected matching degree is then smoothed over time according to the time window of the user's check-in request to obtain the dynamic distribution map of the matching degree. Through spatial distance decay and time smoothing, guest rooms that are closer to the target area and available within the time window receive a higher matching degree, improving the practicality of the recommendation.

[0011] When identifying the candidate room set, this invention sets a matching degree threshold and filters room areas with a matching degree greater than the threshold from the dynamic distribution map of matching degree. For each room in the room area, its current occupancy status is extracted, and rooms that have been booked or are under maintenance are removed. The remaining rooms are sorted from high to low according to their matching degree values, and the top N rooms in the sorting are taken as the initial candidate set. For each room in the initial candidate set, its environmental perception parameters are verified to ensure that they are within the user comfort range. Rooms that pass the verification constitute the candidate room set. This process ensures matching accuracy while taking into account room availability and user comfort.

[0012] When calculating the comprehensive recommendation index, this invention extracts the matching degree value corresponding to the room from the dynamic distribution map of matching degree, as the first recommendation component; obtains the hotel operation strategy weight, which includes the current price discount coefficient, hotel membership level discount coefficient, and hotel inventory pressure coefficient for the room; weights and merges the first recommendation component with the coefficients in the hotel operation strategy weight to obtain the second recommendation component; and correlates and corrects the second recommendation component with the user's historical feedback rating, which is taken from the average of historical guest reviews for the same room type, finally generating the comprehensive recommendation index. Preferably, the current price discount coefficient is specifically calculated by reading the current dynamic pricing strategy for the room from the hotel marketing system and calculating the discount coefficient based on the ratio of the standard price to the current selling price. The comprehensive recommendation index integrates user demand matching degree, hotel operation strategy, and historical reputation, achieving a balance among multiple factors.

[0013] When generating the recommendation list, this invention sorts the comprehensive recommendation index of each room in the candidate room set in descending order of numerical value to obtain an ordered sequence; it adds detailed attribute information to the first K rooms in the ordered sequence, including room photos, prices, and facility descriptions; it adapts the ordered sequence to the screen resolution of the user's terminal to generate a paginated recommendation list; and it pushes the recommendation list to the user's terminal via a wireless communication network, recording user click behavior for subsequent recommendation iterations. This presents the user with intuitive and tailored recommendation results, and continuously optimizes the recommendation model using feedback data.

[0014] Compared with existing technologies, this invention can deeply understand users' implicit preferences by constructing a spatiotemporal sequence of user behavior and a multi-dimensional attribute space. By combining real-time room status data and processing such as spatial distance decay and time smoothing, a dynamic matching distribution map is generated. Based on this, through comprehensive correction of operational strategies and historical evaluations, the recommendation results provide recommendations that ensure user satisfaction while taking into account the hotel's revenue management needs, thus achieving intelligent and accurate recommendations.

[0015] The technical effects and advantages provided by the present invention in the above technical solution are as follows: The spatiotemporal sequence of user behavior is extracted from check-in request information sent by user terminals. This sequence includes timestamps and spatial coordinates of historical check-in records. Specifically, by differential processing and periodic detection of the time intervals between adjacent check-in records, the periodic patterns of user travel are identified. This is combined with the spatial clustering of hotel geographic coordinates to form a behavioral sequence that exhibits both temporal regularity and spatial distribution characteristics. When extracting user preference feature vectors based on this sequence, instead of simply counting the frequency of attribute tags, high-frequency hotels are selected from the spatiotemporally structured sequence, and their attribute tag sets are obtained. The frequency of occurrence is normalized and weighted, and then sparsity correction is applied to obtain a simplified preference feature vector. The constructed multi-dimensional attribute space maps the preference feature vectors to the dimensions of location, price, room type, service, and facilities. Demand intensity curves are generated in each dimension and normalized and aligned to form a continuous multi-dimensional space under a unified coordinate system. When this attribute space is coupled with real-time room status data for analysis, the real-time status data not only includes the spatial layout parameters and facility configuration parameters of the rooms but also environmental sensing parameters such as temperature, humidity, and noise levels collected through an environmental sensor network. During the coupling process, the demand intensity curve of each dimension is matched point-by-point with the actual value of the corresponding dimension of the guest room to obtain the local matching degree. Then, a weighted summation based on dimension weights is used to form the global matching degree. A smoothing correction is applied using a spatial distance decay function and the target occupancy time window, ultimately generating a dynamic matching degree distribution map. This generated distribution map reflects the changes in the matching degree of different guest rooms under user spatiotemporal constraints and diverse needs in real time. It can accurately identify a set of candidate guest rooms that meet user travel cycles and spatial preferences, are available for occupancy, and whose environmental parameters meet comfort requirements. This avoids the recommendation bias caused by traditional methods that ignore spatiotemporal sequences and the real-time multidimensional status of guest rooms. When calculating the comprehensive recommendation index for each guest room in the candidate guest room set, the matching degree value corresponding to that guest room is extracted from the dynamic matching degree distribution map as the first recommendation component. Simultaneously, the hotel operation strategy weight is obtained, which includes the current price discount coefficient, the hotel membership level discount coefficient, and the hotel inventory pressure coefficient. The current price discount coefficient is read from the hotel marketing system's dynamic pricing strategy and calculated based on the ratio of the standard price to the current selling price. The inventory pressure coefficient reflects the guest room vacancy risk and revenue target. The first recommendation component is weighted and fused with the coefficients in the operational strategy weights to obtain the second recommendation component. This second component is then adjusted using the historical average guest reviews for the same room type to generate the final comprehensive recommendation index. The recommendation list generated by sorting according to this index prioritizes highly relevant rooms while dynamically improving the ranking of inventory rooms or promotional rooms that need faster clearance. This ensures that the recommendation results, while meeting individual user needs in terms of time and space, automatically adapt to changes in hotel revenue management, outputting a personalized ranking that balances user satisfaction and hotel operational efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart of an AI-based intelligent hotel room recommendation method. Figure 2 This is a flowchart of the construction of user behavior spatiotemporal sequences and the extraction of user preference feature vectors; Figure 3 This is a flowchart of a room matching and filtering method based on the coupling of multi-dimensional attribute space and real-time room status; Figure 4 It is a periodic autocorrelation analysis graph of the user's historical check-in time interval sequence; Figure 5 This is a scatter plot showing the relationship between the overall recommendation index of candidate guest rooms, matching degree, and operational strategy weights. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 This invention provides an AI-based intelligent hotel room recommendation method, comprising: acquiring check-in request information sent by a user terminal in real time; extracting a spatiotemporal sequence of user behavior from the check-in request information, the spatiotemporal sequence of user behavior including timestamps and spatial coordinates of the user's historical check-in records; extracting user preference feature vectors from the spatiotemporal sequence of user behavior to construct a multi-dimensional attribute space of user needs; acquiring real-time room status data provided by the hotel, the real-time room status data including room spatial layout parameters, facility configuration parameters, and environmental perception parameters; coupling and analyzing the multi-dimensional attribute space with the real-time room status data to generate a dynamic distribution map of the matching degree between rooms and users; identifying a set of candidate rooms that meet user needs based on the dynamic distribution map of the matching degree; calculating a comprehensive recommendation index for each room in the candidate room set, the comprehensive recommendation index being based on the dynamic distribution map of the matching degree and the hotel operation strategy weights; sorting the rooms according to the comprehensive recommendation index from high to low, generating a recommendation list, and sending it to the user terminal. Example 1:

[0020] In specific implementation, please refer to Figure 2 It acquires check-in request information sent by user terminals in real time and extracts the spatiotemporal sequence of user behavior from the check-in request information. The spatiotemporal sequence of user behavior includes the timestamps and spatial coordinates of the user's historical check-in records.

[0021] The method for extracting the spatiotemporal sequence of user behavior is as follows: Parse the user identifier carried in the check-in request information and call the user's historical check-in database associated with that user identifier. The user historical check-in database stores all historical check-in records corresponding to the user identifier, and each historical check-in record contains at least one check-in timestamp field and one hotel geographic coordinate field.

[0022] Each historical check-in record in the user's historical check-in database is arranged in chronological order from earliest to latest based on the value of the check-in timestamp field, forming a sequence of historical check-in records arranged chronologically. The specific values ​​of the check-in timestamps and the hotel's geographical coordinates, including the longitude and latitude values ​​of the hotel's location, are extracted from each historical check-in record in the sequence.

[0023] Differential processing is performed on the check-in timestamps of adjacent historical check-in records to obtain a time interval sequence. The calculation method for differential processing is expressed as follows:

[0024] in, Represents the first time interval in the time interval sequence The value of each time interval. This indicates the sequence of historical check-in records arranged chronologically. The value of the check-in timestamp of each historical check-in record. This indicates the sequence of historical check-in records arranged chronologically. The value of the check-in timestamp of each historical check-in record. The value ranges from to arrive positive integers, This represents the total number of historical check-in records in the historical check-in record sequence. The check-in timestamp value is... Timestamp format, accurate to the second.

[0025] After obtaining the time interval sequence, periodicity detection is performed on the time interval sequence. Periodicity detection uses the autocorrelation function to analyze the time interval sequence and calculate the autocorrelation coefficient sequence. When there is a peak in the autocorrelation coefficient sequence that exceeds a preset autocorrelation threshold, the lag order corresponding to that peak is identified as the periodic pattern of the user's travel. The preset autocorrelation threshold is set to 0.5, and the unit of lag order is days.

[0026] Spatial clustering analysis is performed using hotel geographic coordinates. The method involves extracting the longitude and latitude values ​​of all hotel geographic coordinates from users' historical check-in records, and then using a density-based spatial clustering algorithm to cluster these coordinates, identifying at least one spatial clustering region. The identified user travel cycle patterns and spatial clustering regions are then merged to form a user behavior spatiotemporal sequence. This sequence includes time-ordered cycle pattern markers and spatially ordered spatial clustering characteristic markers.

[0027] Extracting user preference feature vectors from the spatiotemporal sequence of user behavior. Specifically, this involves filtering hotels whose users have checked in more than a preset threshold (3 times). Then, obtaining a set of attribute tags for the filtered hotels, where each attribute tag describes a characteristic of the hotel.

[0028] The frequency of each attribute tag in the statistical attribute tag set is counted in the user's check-in history. The frequency of occurrence is calculated as follows: count the number of user check-in history records containing the attribute tag, and divide the number of records by the total number of user check-in history records. The calculated frequency of occurrence is then normalized using linear normalization, mapping the frequency value to the interval [0,1]. The normalized value is used as the weight value of the attribute tag.

[0029] All attribute tags in the attribute tag set and their corresponding normalized weights are combined to form a multidimensional numerical vector, which is used as the initial feature vector. Sparsity correction is then applied to the initial feature vector. The process involves iterating through all attribute tags in the initial feature vector, calculating the semantic similarity between each pair of attribute tags, and determining that the pair of attribute tags is duplicated or redundant when the semantic similarity exceeds a preset similarity threshold. The attribute tag with the highest weight is retained, and the other attribute tag is deleted. The preset similarity threshold is set to 0.9. Semantic similarity is obtained using cosine similarity based on word vectors. The retained attribute tags and their weights after sparsity correction constitute the user preference feature vector.

[0030] See Figure 4 In the figure, the horizontal axis represents the lag order in days, ranging from 0 to 500 days; the vertical axis represents the autocorrelation coefficient, with a value range of approximately [-0.4, 1.0]. The blue solid line and marked points represent the calculated autocorrelation coefficient sequence, and the red dashed line represents the preset autocorrelation threshold of 0.5.

[0031] The autocorrelation coefficient sequence reaches its maximum value of 1.0 at a lag order of 0, then rapidly declines and exhibits a periodic fluctuation trend. Specifically, the autocorrelation coefficient shows significant peaks at multiple lag orders, such as around day 30, day 60, and day 90, with some peaks exceeding a preset threshold of 0.5, indicating a strong periodic correlation within these time intervals. As the lag order increases, the peak amplitude gradually weakens, and the overall autocorrelation coefficient gradually approaches 0, indicating that the periodic effect gradually diminishes over time.

[0032] Example 2: In practice, the multi-dimensional attribute space for user needs is constructed by mapping user preference feature vectors to a pre-defined hotel attribute classification system. This system consists of five dimensions: location, price, room type, service, and facilities. Each dimension in the hotel attribute classification system contains a set of attribute category identifiers corresponding to that dimension.

[0033] The location dimension includes the following attribute categories: city center area, transportation hub area, commercial area, scenic area, and residential area. The price dimension includes the following attribute categories: economy, comfort, luxury, and deluxe. Each price range corresponds to a numerical range: economy (less than 300 RMB per day), comfort (300-600 RMB per day), luxury (600-1200 RMB per day), and deluxe (more than 1200 RMB per day). The room type dimension includes the following attribute categories: standard king room, standard twin room, business suite, executive suite, and family suite. The service dimension includes the following attribute categories: airport transfer, laundry service, catering service, meeting service, and butler service. The facilities dimension includes the following attribute categories: gym facilities, swimming pool facilities, parking facilities, Wi-Fi facilities, and smart home facilities.

[0034] Each attribute label in the user preference feature vector is matched with an attribute category identifier from one of the five dimensions in the hotel attribute classification system. The matching method is semantic matching. The edit distance similarity is calculated between the text content of the attribute label and the text content of the attribute category identifier. When the edit distance similarity exceeds a preset edit distance threshold, the attribute label is considered to have matched the attribute category identifier successfully. The preset edit distance threshold is set to 0.8. After a successful match, the weight value corresponding to that attribute label in the user preference feature vector is assigned to the corresponding attribute category identifier in the corresponding dimension of the hotel attribute classification system.

[0035] Within each dimension, a demand intensity curve is generated based on the weight values ​​assigned to each attribute category identifier within that dimension. The demand intensity curve is generated as follows: all attribute category identifiers within that dimension are treated as discrete nodes on the curve. The horizontal axis is plotted using the ranking number of the attribute category identifiers, and the vertical axis is plotted using the assigned weight values. Interpolation smoothing is then applied to the discrete nodes to form a continuous demand intensity curve. The ranking number generation rule is as follows: according to the preset ranking order of the attribute category identifiers within that dimension in the hotel attribute classification system, integer ranking numbers are assigned sequentially, starting from 1.

[0036] The demand intensity curves for each dimension are normalized and aligned. The normalization and alignment process is as follows: obtain the ordinate values ​​of all discrete nodes on the demand intensity curve for each of the five dimensions, and scale the ordinate values ​​on the demand intensity curve for each dimension. The scaling method is to divide the ordinate value of each discrete node by the maximum value of the ordinate values ​​of all discrete nodes on that demand intensity curve, so that the ordinate values ​​on the demand intensity curves of the five dimensions are uniformly mapped to the interval [0,1], forming a multi-dimensional attribute space under a unified coordinate system.

[0037] Interpolation is used to fill blank areas in the multi-dimensional attribute space. Blank areas refer to locations in the multi-dimensional attribute space where no data points exist between two known data points. Linear interpolation is used for filling, and the interpolation calculation is based on the following expression:

[0038] in, Indicated on the x-axis The ordinate value is obtained by interpolation and filling. Represents the x-axis The x-coordinate of the nearest known data point on the left. Represents the x-axis The ordinate value of the nearest known data point on the left. Represents the x-axis The x-coordinate of the nearest known data point on the right. Represents the x-axis The ordinate value of the nearest known data point on the right. The x-coordinate value corresponding to the blank area that needs to be interpolated and filled. The value is located in and Between. Using linear interpolation, interpolation calculations are performed on all identified blank areas in the multi-dimensional attribute space to obtain the result on the horizontal axis. Fill y-coordinate value at the location This ensures that the demand intensity curve in each dimension is continuous and differentiable in the multi-dimensional attribute space under a unified coordinate system.

[0039] Example 3: In practice, real-time guest room status data is obtained from the hotel. This data includes spatial layout parameters, facility configuration parameters, and environmental perception parameters.

[0040] The hotel management system interface collects real-time data on the current occupancy, cleanliness, and maintenance status of each guest room. The hotel management system interface uses an HTTP-based application programming interface (API) with JSON data exchange format. The request method is a GET request, with the request URL including the hotel's unique identifier and the guest room number as path parameters. Upon receiving the request, the hotel management system interface returns a response containing three status fields: occupancy status, cleanliness status, and maintenance status. The occupancy status field is represented as "occupied" or "available," the cleanliness status field as "cleaned" or "not cleaned," and the maintenance status field as "under maintenance" or "normal." The data collection process involves iterating through all guest room numbers, constructing a separate GET request for each guest room number, sending the request to the hotel management system interface, receiving the response data, parsing the values ​​of the occupancy, cleanliness, and maintenance status fields from the response data, and associating these three status field values ​​with the corresponding guest room number for storage.

[0041] The spatial layout parameters for each guest room are extracted from the hotel's spatial database. The hotel spatial database is a relational database, internally storing a spatial layout data table. Each row in the spatial layout data table corresponds to one guest room. The columns of the spatial layout data table include guest room number, room orientation, floor height, and area. The room orientation column takes one of the following values: East, South, West, North, Southeast, Northeast, Southwest, or Northwest. The floor height column takes a value in meters, and the area column takes a value in square meters. The extraction process is as follows: A structured query statement is sent to the hotel spatial database. The structured query statement includes a SELECT clause, specifying the guest room number, room orientation, floor height, and area columns. It also includes a FROM clause, specifying the name of the spatial layout data table. The hotel spatial database executes the structured query statement, returning a result set. Each row in the result set contains a guest room number, a room orientation value, a floor height value, and an area value. Each row in the result set is stored as a spatial layout parameter for one guest room.

[0042] The facility configuration parameters for each guest room are extracted from the hotel facility database. The hotel facility database is a relational database, internally storing a facility configuration data table. Each row in the facility configuration data table corresponds to one guest room. The columns of the facility configuration data table include a guest room number column, a bed type column, a bathroom equipment column, and a smart home device list column. The bed type column takes one of the following values: single bed, double bed, or king bed. The bathroom equipment column takes a combination identifier containing a bathtub and a shower. The smart home device list column takes a comma-separated string of device names, and the devices appearing in the device name string include at least one of the following: smart lighting, smart curtains, smart air conditioning, and smart door locks. The extraction process involves sending a structured query statement to the hotel facility database. The structured query statement includes a SELECT clause, which specifies the guest room number, bed type, bathroom equipment, and smart home device list columns. The structured query statement also includes a FROM clause, which specifies the name of the facility configuration data table. The hotel facilities database executes a structured query, returning a result set. Each row in the result set contains a room number, a bed type value, a bathroom fixture value, and a list of smart home devices. Each row in the result set is stored as a facility configuration parameter for a guest room.

[0043] Environmental sensing parameters for each guest room are collected from an environmental sensor network. This network consists of temperature sensor nodes, humidity sensor nodes, and noise sensor nodes deployed in each guest room. Each temperature sensor node collects the temperature value of its guest room at a fixed interval of 60 seconds. Each humidity sensor node collects the humidity value of its guest room at a fixed interval of 60 seconds. Each noise sensor node collects the noise level value of its guest room at a fixed interval of 60 seconds. The temperature, humidity, and noise sensor nodes transmit the collected data to the environmental sensor network's data aggregation gateway via a wireless communication protocol. The data aggregation gateway then forwards the received data to the data storage server.

[0044] The data storage server stores an environmental sensing parameter data table. Each row in the table corresponds to the data collected by a sensor node at a single acquisition moment. The table includes columns for room number, acquisition timestamp, temperature, humidity, and noise level. Temperature values ​​are expressed in degrees Celsius, humidity values ​​in percentages, and noise level values ​​in decibels. The acquisition process involves sending a data query request to the data storage server, specifying a query time window. This time window is a fixed time range, 300 seconds, preceding the current moment. After receiving a data query request, the data storage server retrieves all records from the environmental sensing parameter data table whose collection timestamp values ​​fall within the query time window. It then groups all retrieved records according to the room number column. For each room number, the server calculates the arithmetic mean of all values ​​in the temperature column within that group to obtain the temperature parameter value for that room number; similarly, it calculates the arithmetic mean of all values ​​in the humidity column within that group to obtain the humidity parameter value; and finally, it calculates the arithmetic mean of all values ​​in the noise level column within that group to obtain the noise parameter value for that room number.

[0045] The arithmetic mean is calculated as follows:

[0046] in, This represents the arithmetic mean of a certain type of environmental perception parameter within a group corresponding to a room number over the query time window. This represents the total number of records of this type of environmental perception parameter collected within the group corresponding to a guest room number. This represents the first environmental perception parameter of this type collected within the group corresponding to a guest room number. The value of each numerical record. The value is from arrive A positive integer. The temperature, humidity, and noise parameter values ​​corresponding to each room number are combined and stored as the environmental perception parameters for that room number.

[0047] Example 4: In specific implementation, please refer to Figure 3This study couples a multi-dimensional attribute space with real-time guest room status data for analysis. The multi-dimensional attribute space includes demand intensity curves across five dimensions: location, price, room type, service, and facilities. The real-time guest room status data includes spatial layout parameters, facility configuration parameters, and environmental perception parameters for each guest room.

[0048] For each dimension of the demand intensity curve in the multi-dimensional attribute space, a point-by-point matching is performed with the corresponding actual value in the real-time guest room status data to obtain the local matching degree for each dimension. The point-by-point matching is implemented as follows: For the location dimension, the geographical coordinates of the hotel where the guest room is located are extracted from the spatial layout parameters of the real-time guest room status data. The hotel's geographical coordinates are input into the demand intensity curve of the location dimension. The horizontal axis of the demand intensity curve of the location dimension represents the spatial division area identifier of the geographical coordinates. The hotel's geographical coordinates are mapped to the corresponding spatial division area identifier, and the vertical axis value corresponding to this spatial division area identifier on the demand intensity curve is read as the local matching degree of the location dimension. For the price dimension, the current selling price of the guest room is obtained from the hotel marketing system associated with the facility configuration parameters of the real-time guest room status data. The current selling price is input into the demand intensity curve of the price dimension. The horizontal axis of the demand intensity curve of the price dimension represents the price value. A linear interpolation method is used to determine the vertical axis value corresponding to the current selling price on the demand intensity curve as the local matching degree of the price dimension. For the room type dimension, bed type values ​​are extracted from the facility configuration parameters of the real-time guest room status data. These values ​​are then input into the demand intensity curve for the room type dimension, where the horizontal axis represents the room type category identifier. The vertical axis value corresponding to this identifier is used as the local matching degree for the room type dimension. For the service dimension, a list of available services is retrieved from the hotel service database associated with the facility configuration parameters of the real-time guest room status data. Each service item in this list is mapped to a horizontal axis point on the demand intensity curve for the service dimension. The arithmetic mean of the vertical axis values ​​for all services in the list is calculated as the local matching degree for the service dimension. For the facility dimension, a list of smart home devices is extracted from the facility configuration parameters of the real-time guest room status data. Each device name in this list is mapped to a horizontal axis point on the demand intensity curve for the facility dimension. The arithmetic mean of the vertical axis values ​​for all devices in the list is calculated as the local matching degree for the facility dimension.

[0049] Obtain a preset set of dimension weight coefficients, containing five dimension weight coefficients: location, price, room type, service, and facilities. The values ​​of all five dimension weight coefficients are within the interval [0,1], and the sum of the five dimension weight coefficients equals 1. The location weight coefficient is set to 0.3, the price weight coefficient to 0.25, the room type weight coefficient to 0.2, the service weight coefficient to 0.15, and the facilities weight coefficient to 0.1. These dimension weight coefficient values ​​are determined based on the statistical distribution of the importance of user room selection dimensions in historical booking data. Multiply the local matching degree of each of the five dimensions by its corresponding dimension weight coefficient, and then sum the five products to obtain the global matching degree.

[0050] A spatial distance decay function is constructed centered on the target geographic location carried in the user's check-in request. The target geographic location is represented by longitude and latitude values. The hotel geographic coordinates corresponding to each room are extracted from the spatial layout parameters of the real-time room status data. For each room, the Euclidean distance between the hotel geographic coordinates and the target geographic location is calculated. The spatial distance decay function adopts an exponential decay form, and the global matching degree is multiplied by the spatial distance decay coefficient to obtain the spatially corrected matching degree. The calculation method of the spatially corrected matching degree is expressed as follows:

[0051] in, This indicates the matching degree after space correction. Indicates the global matching degree. Represented by natural constant An exponential function with base 0. Indicates the spatial distance attenuation coefficient. This represents the Euclidean distance between the hotel's geographical coordinates and the target geographical location, expressed in kilometers. Spatial distance attenuation coefficient. The value is set to 0.05. The basis for this value is that when the Euclidean distance between the hotel's geographical coordinates and the target geographical location increases by 10 kilometers, the matching degree decreases by about 39%. This rate of decrease is consistent with the statistical experience value of users' travel distance sensitivity.

[0052] The spatially corrected matching degree is smoothed over time according to the user's check-in request time window to obtain a dynamic distribution map of the matching degree. The user's check-in request carries the check-in date and check-out date, and the time window is defined as a continuous set of dates from the check-in date to the check-out date. For each date within the time window, the above operations of obtaining real-time room status data and calculating the spatially corrected matching degree are performed to obtain the spatially corrected matching degree for each day. A moving average is calculated for the spatially corrected matching degree of all dates within the time window. The window length of the moving average is the same as the total number of days included in the time window. The result of the moving average is used as the matching degree value of that room within that time window in the dynamic distribution map of the matching degree. The matching degree values ​​of all rooms are represented by color intensity on a plane coordinate system to form the dynamic distribution map of the matching degree.

[0053] After obtaining the dynamic distribution map of matching scores, a set of candidate guest rooms that meet user needs is identified. A matching score threshold is set to 0.7, based on the principle of retaining guest room areas with matching scores in the top 30%. Guest room areas with matching scores greater than the matching score threshold are then selected from the dynamic distribution map.

[0054] For each room within the selected guest room area, extract the current occupancy and maintenance status from the real-time room status data. Remove booked rooms, defined as those with the current occupancy status field showing "occupied." Remove rooms under maintenance, defined as those with the maintenance status field showing "under maintenance." The remaining rooms after these removals constitute the remaining room set.

[0055] For each room in the remaining room set, extract its matching degree value from the matching degree dynamic distribution map, and sort them from high to low matching degree values. Select the top N rooms in the sorted list as the initial candidate set, where N is set to 10, based on the maximum number of rooms displayed in the single-page recommendation list.

[0056] For each guest room in the initial candidate set, temperature, humidity, and noise parameters are extracted from the environmental sensing parameters of the guest room's real-time status data. User comfort ranges are pre-defined, including temperature, humidity, and noise comfort ranges. The temperature comfort range is set to 20°C to 26°C, the humidity comfort range to 40% to 60%, and the noise comfort range to less than 45 decibels. The system verifies whether the temperature, humidity, and noise parameters of each guest room in the initial candidate set fall within their respective comfort ranges. Guest rooms whose temperature, humidity, and noise parameters all fall within their respective comfort ranges are approved, and these approved guest rooms constitute the candidate guest room set.

[0057] Example 5: In practice, a comprehensive recommendation index is calculated for each room in the candidate room set. The comprehensive recommendation index is based on a dynamic distribution map of matching degree and the weights of hotel operation strategies.

[0058] The matching degree values ​​corresponding to candidate rooms are extracted from the matching degree dynamic distribution map. The matching degree dynamic distribution map is a data structure indexed by room number, storing the matching degree value of each room after time smoothing within a time window. For each room in the candidate room set, the room number is used as the query key to query the corresponding matching degree value in the matching degree dynamic distribution map, and the queried matching degree value is used as the first recommendation component.

[0059] Obtain the hotel operation strategy weight. The hotel operation strategy weight consists of three coefficients: current price discount coefficient, hotel membership level discount coefficient, and hotel inventory pressure coefficient.

[0060] The current price discount factor is obtained by retrieving the current dynamic pricing strategy of the candidate rooms from the hotel marketing system. The hotel marketing system is a relational database that stores pricing information for each room. It contains a pricing data table with columns for room number, standard price, and current selling price. The standard price column represents the listed price of the room without any discounts, in yuan. The current selling price column represents the actual selling price of the room at the current moment, in yuan. A structured query statement is sent to the hotel marketing system. This query statement includes a SELECT clause specifying the standard price and current selling price columns, and a WHERE clause specifying that the room number column equals the room number of the candidate room. The hotel marketing system executes the structured query statement and returns a row containing the standard price and current selling price values. The current selling price is divided by the standard price, and the calculated ratio is used as the current price discount factor. The current price discount factor ranges from (0,1), with a smaller factor indicating a larger discount.

[0061] The method for obtaining hotel membership level discount coefficients is as follows: Parse the user identifier carried in the user's check-in request information and call the user membership management database. The user membership management database stores a user identifier column and a membership level column, where the membership level value is one of the following: non-member, silver member, gold member, or diamond member. A predefined hotel membership level discount coefficient mapping table records the discount coefficient value corresponding to each membership level. The discount coefficient value for non-members is 1.00, for silver members it is 0.95, for gold members it is 0.90, and for diamond members it is 0.85. These discount coefficient values ​​are set based on the price discount ratios for each level in the hotel membership system. The user membership management database is queried for the membership level value corresponding to the user identifier, and the discount coefficient value corresponding to that membership level value is read from the hotel membership level discount coefficient mapping table as the hotel membership level discount coefficient.

[0062] The hotel inventory pressure coefficient is obtained by querying the overall room occupancy rate of the hotel to which the candidate rooms belong through the hotel management system's interface. The hotel management system's interface provides an inventory statistics interface. The request URL for the inventory statistics interface includes the hotel's unique identifier as a path parameter, and the request method is a GET request. After receiving the inventory statistics request, the hotel management system's interface returns response data containing two fields: the total number of hotel rooms and the number of occupied rooms. The total number of hotel rooms represents the total number of rooms available for sale, and the number of occupied rooms represents the total number of rooms currently booked or occupied. The overall room occupancy rate is calculated by dividing the value of the occupied room field by the value of the total number of hotel rooms. A predefined piecewise function for the hotel inventory pressure coefficient is used: when the overall room occupancy rate is less than 0.5, the hotel inventory pressure coefficient is 1.00; when the overall room occupancy rate is greater than or equal to 0.5 and less than 0.8, the hotel inventory pressure coefficient is 0.95; and when the overall room occupancy rate is greater than or equal to 0.8, the hotel inventory pressure coefficient is 0.85. The aforementioned segmented thresholds and coefficient values ​​are based on the hotel revenue management strategy's arrangements for price reductions and promotions at different occupancy levels. The corresponding coefficient value is retrieved from the segmented function of the hotel inventory pressure coefficient based on the calculated overall room occupancy rate.

[0063] The first recommendation component is weighted and fused with the three coefficients in the hotel operation strategy weighting to obtain the second recommendation component. The weighted fusion is calculated as follows: multiply the current price discount coefficient, the hotel membership level discount coefficient, and the hotel inventory pressure coefficient to obtain a fusion coefficient value; multiply the first recommendation component by the fusion coefficient value, and the product is used as the second recommendation component.

[0064] The second recommendation component is correlated and corrected with historical user feedback ratings to generate a comprehensive recommendation index. Historical user feedback ratings are taken from the average of historical guest reviews for the same room type. The historical user feedback ratings are obtained by extracting the bed type value from the facility configuration parameters of the candidate room's real-time status data, using this bed type value as the room type identifier. The hotel review database is accessed; it stores review data tables with columns for room type identifier, user identifier, and rating value. The rating value is an integer from 1 to 5. A structured query statement is sent to the hotel review database. This statement includes a SELECT clause specifying the rating value column and a WHERE clause specifying that the room type identifier is equal to the extracted bed type value. The hotel review database executes the structured query, returning a set of rating values ​​containing all historical user rating values ​​corresponding to that room type identifier. The arithmetic mean of all ratings in the rating set is calculated, and then normalized by dividing the arithmetic mean by 5. The normalized user historical feedback rating ranges from 0.2 to 1. The comprehensive recommendation index is calculated as follows:

[0065] in, This represents the overall recommendation index. This indicates the second recommended component. This represents the normalized user history feedback rating. Indicates the historical feedback influence factor. This represents the basic bias term. The value is set to 0.6. The value is set to 0.4. and The sum is 1. The value of 0.6 is based on assigning a 60% correction weight to historical feedback scores, maintaining a significant proportion of user evaluation factors in the recommendation results; The value of 0.4 is chosen to provide a base rating bias of 0.4 when historical feedback rating data is lacking.

[0066] A recommendation list is generated and sent to the user terminal. The overall recommendation index of each room in the candidate room set is sorted in descending order of value. The descending order uses a comparison sorting algorithm, placing the room with the highest overall recommendation index at the first position of the sequence, the room with the second highest overall recommendation index at the second position, and so on, to obtain an ordered sequence. Detailed attribute information is added to the first K rooms in the ordered sequence, with K set to 5, based on the number of room cards displayed in the single-page recommendation list on the user terminal. Detailed attribute information includes room photos, prices, and facility descriptions. Room photos are obtained from the hotel room image database, which stores the file paths of room photos indexed by room number; the corresponding room photo file is read according to the room number. Prices are read from the current selling price column of the hotel marketing system. Facility descriptions are extracted from the facility configuration parameters of the room's real-time status data, including bed type values, bathroom equipment values, and a list of smart home devices, and these values ​​are concatenated into a text string as the facility description.

[0067] The ordered sequence is adapted to the user terminal's screen resolution to generate a paginated recommendation list. The user terminal's screen resolution is obtained from the device model field carried in the check-in request information sent by the user terminal. Screen resolution is represented as width in pixels multiplied by height in pixels. When the screen resolution's width in pixels is less than or equal to 720, 3 rooms are recommended per page; when the screen resolution's width in pixels is greater than 720, 5 rooms are recommended per page. Based on the number of rooms displayed per page, the ordered sequence is split into multiple pages. Each page contains no more than the number of rooms displayed on that page, along with their additional detailed attribute information, forming a paginated recommendation list.

[0068] The recommendation list is pushed to the user terminal via a wireless communication network. The wireless communication network uses one of four types: fourth-generation mobile communication network, fifth-generation mobile communication network, or wireless local area network. The push uses a POST request method based on the HTTP protocol, serializing the paginated recommendation list into JSON format data and placing it in the request body of the POST request. After receiving the POST request, the user terminal parses the JSON format data and renders and displays the first page of the recommendation list. Simultaneously, a click event listener is embedded on each room card in the recommendation list. When a user clicks any room card, the user terminal transmits the corresponding room number and user identifier back to the server via the wireless communication network. The server stores the click event record in a user behavior log database, which contains columns for user identifier, room number, click timestamp, and event type. The event type column is set to "recommended click" for subsequent recommendation iterations.

[0069] See Figure 5 In the graph, the horizontal axis represents the first recommendation component of the candidate rooms, which is the numerical value of the matching degree based on the dynamic distribution map of matching degree, ranging from 0.2 to 1.0. The vertical axis represents the comprehensive recommendation index of the corresponding room, ranging from approximately 0.1 to 1.0. Each scatter point in the graph represents a candidate room, and the color of the scatter point maps to the corresponding fusion coefficient (price discount coefficient × member discount coefficient × inventory pressure coefficient). The color bar shows that the value of this fusion coefficient ranges from 0.58 to 1.00, with the color gradient from purple to yellow. The more yellow the color, the higher the fusion coefficient.

[0070] Overall, the scatter plots are distributed below the diagonal reference line (y=x), indicating that the comprehensive recommendation index is generally lower than the corresponding first recommendation component, which is consistent with the calculation logic of the comprehensive recommendation index in Example 5 after weighted fusion and historical feedback correction. As the first recommendation component increases, the comprehensive recommendation index shows a positive correlation upward trend, indicating that rooms with higher matching degree generally have higher comprehensive recommendation indices after correction by operational strategy weights and user historical feedback scores.

[0071] In terms of color distribution, the scatter points with higher fusion coefficients (yellow) are mainly concentrated in areas with higher first-recommendation components and higher overall recommendation indices. This indicates that among rooms with high matching degrees, those enjoying larger price discounts, higher membership benefits, and lower inventory pressure are more likely to obtain higher overall recommendation indices. Conversely, the scatter points with lower fusion coefficients (purple) are more distributed in the lower part of the graph, showing that their overall recommendation indices are more affected by the weighting of operational strategies.

[0072] The densely packed areas in the graph are concentrated between 0.5 and 0.9 for the first recommendation component, and the comprehensive recommendation index is between 0.25 and 0.65, reflecting the actual distribution range of candidate rooms in terms of matching degree and comprehensive recommendation index. Rooms within this range are the main candidate rooms for generating the recommendation list, demonstrating the effective distinguishing role of the comprehensive recommendation index in Example 5.

[0073] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A hotel room intelligent recommendation method based on artificial intelligence, characterized in that, include: The system acquires check-in request information sent by user terminals in real time, and extracts a spatiotemporal sequence of user behavior from the check-in request information. The spatiotemporal sequence of user behavior includes timestamps and spatial coordinates of the user's historical check-in records. User preference feature vectors are extracted from the spatiotemporal sequence of user behavior to construct a multi-dimensional attribute space of user needs; Acquire real-time status data of guest rooms provided by the hotel, including spatial layout parameters, facility configuration parameters, and environmental perception parameters of the guest rooms; The multi-dimensional attribute space is coupled and analyzed with the real-time status data of the guest rooms to generate a dynamic distribution map of the matching degree between guest rooms and users. Based on the dynamic distribution map of matching degree, a set of candidate guest rooms that meet the user's needs is identified; For each room in the candidate room set, a comprehensive recommendation index is calculated, which is based on the matching degree dynamic distribution map and the hotel operation strategy weights. The system sorts users by their overall recommendation index from highest to lowest, generates a recommendation list, and sends it to the user's device.

2. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The step of extracting the spatiotemporal sequence of user behavior from the check-in request information includes: Parse the user identifier in the check-in request information and retrieve the user's historical check-in database; Arrange each check-in record in the historical check-in database in chronological order, and extract the check-in timestamp and hotel geographical coordinates corresponding to each record; The time intervals between adjacent check-in records are differentially processed to obtain a time interval sequence; Periodic detection is performed on the time interval sequence to identify the periodic pattern of user travel, and combined with the spatial clustering of the hotel's geographical coordinates to form the spatiotemporal sequence of user behavior.

3. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The step of extracting user preference feature vectors from the spatiotemporal sequence of user behavior includes: From the spatiotemporal sequence of user behavior, select hotels whose number of stays exceeds a preset threshold, and obtain the attribute tag set of these hotels; The frequency of each attribute tag in the user's check-in history is counted, and the normalized frequency is used as the weight value of that attribute tag. The attribute label set and its corresponding weight values ​​are combined into an initial feature vector; The initial feature vector is subjected to sparsity correction, and duplicate or redundant attribute labels are removed to obtain the user preference feature vector.

4. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The construction of the multi-dimensional attribute space for user needs includes: The user preference feature vector is mapped to a preset hotel attribute classification system, which includes location, price, room type, service, and facilities dimensions. For each dimension, a demand intensity curve for that dimension is generated based on the weight values ​​in the user preference feature vector; The demand intensity curves of each dimension are normalized and aligned to form a multi-dimensional attribute space under a unified coordinate system. The blank areas in the multi-dimensional attribute space are filled by interpolation to ensure that the demand intensity curve in each dimension is continuous and differentiable.

5. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The acquisition of real-time room status data provided by the hotel includes: The hotel management system interface is used to collect the current occupancy status, cleaning status, and maintenance status of each guest room in real time. Extract the spatial layout parameters of each guest room from the hotel space database. The spatial layout parameters include room orientation, floor height, and area. Extract the facility configuration parameters for each guest room from the hotel facilities database. The facility configuration parameters include a list of bed type, bathroom facilities, and smart home devices. Environmental sensing parameters for each guest room are collected from an environmental sensor network, including temperature, humidity, and noise levels.

6. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The coupling analysis of the multi-dimensional attribute space with the real-time status data of the guest rooms includes: The demand intensity curve of each dimension in the multi-dimensional attribute space is matched point by point with the actual value of the corresponding dimension in the real-time status data of the guest room to obtain the local matching degree of each dimension. Based on the preset dimension weight coefficients, the local matching degree of each dimension is weighted and summed to obtain the global matching degree; Using the target geographical location in the user's check-in request as the center, a spatial distance decay function is constructed. The global matching degree is multiplied by the spatial distance decay coefficient to obtain the spatially corrected matching degree. The spatially corrected matching degree is smoothed over time according to the time window of the user's check-in request to obtain the dynamic distribution map of the matching degree.

7. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The step of identifying a set of candidate guest rooms that meet user needs based on the matching degree dynamic distribution map includes: Set a matching threshold, and filter out guest room areas with a matching degree greater than the matching threshold from the matching degree dynamic distribution map; For each guest room in the guest room area, extract its current occupancy status and remove rooms that have been booked or are under maintenance; Sort the remaining guest rooms in descending order of matching degree value, and take the top N guest rooms as the initial candidate set; For each guest room in the preliminary candidate set, verify whether its environmental perception parameters are within the user comfort range. The guest rooms that pass the verification constitute the candidate guest room set.

8. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The calculation of the comprehensive recommendation index for each room in the candidate room set includes: The matching degree value corresponding to the guest room is extracted from the matching degree dynamic distribution map and used as the first recommendation component; Obtain the hotel operation strategy weight, which includes the current price discount coefficient of the room, the hotel membership level discount coefficient, and the hotel inventory pressure coefficient; The first recommendation component is weighted and fused with each coefficient in the hotel operation strategy weight to obtain the second recommendation component; The second recommendation component is correlated and corrected with the user's historical feedback rating, which is taken from the average historical resident evaluation of the same room type, to generate the comprehensive recommendation index.

9. The intelligent hotel room recommendation method based on artificial intelligence according to claim 1, characterized in that, The step of sorting users by comprehensive recommendation index from high to low, generating a recommendation list, and sending it to the user terminal includes: The comprehensive recommendation index of each room in the candidate room set is sorted in descending order of numerical value to obtain an ordered sequence; Detailed attribute information is added to the first K rooms in the ordered sequence, including room photos, prices, and facility descriptions. The ordered sequence is adapted to the screen resolution of the user terminal to generate a paginated recommendation list; The recommendation list is pushed to the user's terminal via a wireless communication network, and the user's click behavior is recorded for subsequent recommendation iterations.

10. The intelligent hotel room recommendation method based on artificial intelligence according to claim 8, characterized in that, The process of obtaining the current price discount coefficient in the hotel operation strategy weighting is as follows: read the current dynamic pricing strategy of the room from the hotel marketing system, and calculate the discount coefficient based on the ratio of the standard price to the current selling price.

Images