CN117501286A - Hotel demand model based on artificial intelligence - Google Patents

Abstract

Embodiments generate a demand model for potential hotel clients of a hotel room. Embodiments form a plurality of clusters based on characteristics of potential hotel customers, each cluster including a corresponding weight and cluster probability. An embodiment generates an initial estimated mixture of multiple logic ("MNL") models corresponding to each of a plurality of clusters, the mixture of MNL models including weighted likelihood functions based on features and weights. The embodiment determines the corrected cluster probability and updates the weights. Embodiments estimate a blend of updated estimates of the MNL model based on the modified cluster probabilities and the updated weights and maximize the weighted likelihood function. Based on the updated estimated mix of updated weights and MNL models, embodiments generate a demand model adapted to predict a selection probability of a combination of room categories and rate codes for potential hotel customers.

Description

Hotel demand model based on artificial intelligence

Cross Reference to Related Applications

The present application claims priority from U.S. provisional patent application Ser. No. 63/215,688, filed on 6/28 of 2021, the disclosure of which is incorporated herein by reference.

Technical Field

One embodiment relates generally to computer systems, and in particular to computer systems that generate artificial intelligence based hotel (hotel) demand models.

Background

The increased competition in the hotel industry has led hotel operators to seek more innovative revenue management strategies, such as personalized pricing and recommendations. Over the past few years, hotel operators have begun to understand that not all guests are the same, and traditional universal (one-size-fit-all) strategies may prove ineffective. Therefore, hotels need to profile their guests and offer the guests the appropriate products/services at the appropriate price to achieve the goal of maximizing the profits of the hotel.

Disclosure of Invention

Embodiments generate a demand model for potential hotel clients of a hotel room. Embodiments form a plurality of clusters based on characteristics of potential hotel customers, each cluster including a corresponding weight and cluster probability. An embodiment generates an initial estimated mixture of multiple logic ("MNL") models corresponding to each of a plurality of clusters, the mixture of MNL models including weighted likelihood functions based on features and weights. The embodiment determines the corrected cluster probability and updates the weights. Embodiments estimate a blend of updated estimates of the MNL model based on the modified cluster probabilities and the updated weights and maximize the weighted likelihood function. Based on the updated estimated mix of updated weights and MNL models, embodiments generate a demand model adapted to predict a selection probability of a combination of room categories and rate codes for potential hotel customers.

Drawings

Fig. 1 is an overview block diagram of a hotel reservation system in accordance with an embodiment of the invention.

Fig. 2 is a block diagram of a computer server/system according to an embodiment of the invention.

FIG. 3 is a flow diagram of the functionality of the room demand model module of FIG. 2 for generating a room demand model, according to one embodiment.

FIG. 4 illustrates an example of initial clustering according to an embodiment.

Fig. 5 is an example illustrating various offered prices, room categories, and rate codes.

FIG. 6 illustrates selection modeling of guest clusters according to an example embodiment.

Fig. 7 illustrates initial assignment of MNL models to each cluster according to an embodiment.

Fig. 8 illustrates a proposed likelihood function for use with EM functions according to an embodiment.

Fig. 9 illustrates a portion of EM functionality according to an embodiment.

Fig. 10 illustrates a portion of EM functionality according to an embodiment.

Fig. 11-16 illustrate examples of embodiments of the invention for three clusters.

FIG. 17 illustrates a comparison of prediction accuracy with iteration between CCR and MSE according to embodiments of the present invention.

Fig. 18 illustrates how the cluster characteristics change with iteration given two clusters, according to an embodiment.

Detailed Description

Embodiments predict customer selection of hotel room categories and associated service types based on estimated parameters of discrete selection models built on observed, dynamically determined clusters. Each observation corresponds to a selection made by a customer subscribing to a room in a hotel and selecting an associated service type from an ordered set of room category and service type pairs offered at a particular price. Each room category and service type is described by a feature set that determines the value or utility of the customer's selection. Furthermore, each client is characterized by its own set of attributes that determine the cluster to which the client belongs, also known as the "persona (persona) type. It is assumed that each role type may have its own utility of subscription selection.

The selection probability is modeled as a multiple-term logic function based on the room-service pair utility for each role type. Embodiments improve the accuracy of predictions and build a basis for normative analysis applications that optimize personalized offers by maximizing expected revenue. Embodiments may be used as stand-alone systems or as a central part of a personalized price optimization system for personalized hotel rooms and a display optimization system for the order of room categories and rate codes. Rather than using static clustering traditionally used for this purpose, embodiments utilize iterative reconfigurable dynamic clustering based on semi-parametric blending of discrete selection models to adequately reflect the customer's selection behavior.

In general, in the hotel industry, and other comparative industries, increased competition is pushing more innovative revenue management practices, such as personalized offers and pricing. Not all customers are identical and traditional generic policies may prove ineffective. Accurate estimation of demand is critical as an input to personalized recommendation systems.

Embodiments address the need for a more accurate estimation of hotel room demand by modeling demand to account for heterogeneous customers having different characteristics: (1) Willingness to pay (indicated by the selected price range); (2) Rate plan selections (enterprise discounts, including breakfast, etc.); (3) travel attributes; (4) booking channels; (5) a subscription window; (6) residence time; (7) arrival date; and/or (8) community/family size, number of children, etc. Factors affecting the selection may include room characteristics, rate plan characteristics, prices, and the order in which the offers are displayed.

Reference will now be made in detail to embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. It will be apparent, however, to one of ordinary skill in the art that the present disclosure may be practiced without these specific details. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the embodiments. Wherever possible, the same reference numbers will be used for the same elements.

Fig. 1 is an overview block diagram of a hotel reservation system 100 in accordance with an embodiment of the present invention. Fig. 1 includes a reservation channel 102 with which potential hotel clients can interact to reserve hotel rooms. These channels include global distribution system ("GDS") 111, including "Amadeus", "Sabre", "Travel Port", etc.; online travel agency ("OTA") 112, including "booking. Com", "expect", and the like; a meta search site 113; and any other means by which a customer subscribes to a hotel room, including websites maintained by a hotel chain or individual hotels.

Each hotel chain operation 104 is accessed by an application programming interface ("API") 140 as a Web service, such as a "WebLogic server" from Oracle corporation. Hotel chain management 104 includes a hotel property management system ("PMS") 121 (such as "OPERA cloud property management" from Oracle corporation), a hotel central reservation system ("CRS") 122, and a demand modeling module 150 that interfaces with systems 121 and 122 to provide optimized demand modeling as disclosed herein.

Hotel customers or potential hotel customers who use the system 100 to obtain a hotel room typically participate in a three-stage booking process. A region availability search is first conducted. A plurality of hotel chains are shown and hotel CRSs 122 provide static data. Static data may include minimum/maximum rates, available dates, etc.

If the reservation customer selects a hotel, they go to the next step, a property search, including a single hotel property, multiple rooms, and rate plans. For a single hotel property, the information may include room category description data, rate plan description, and room price, each of which is displayed in a particular order. Property searches include real-time availability data and results of reservation clients selecting rooms. Once the room is selected, the final step is to make the final reservation and secure the reservation through a credit card or other form of payment.

Fig. 2 is a block diagram of a computer server/system 10 according to an embodiment of the invention. Although shown as a single system, the functionality of system 10 may be implemented as a distributed system. Further, the functionality disclosed herein may be implemented on separate servers or devices that may be coupled together through a network. Further, one or more components of system 10 may not be included. For example, when implemented as web server or cloud-based functions, system 10 is implemented as one or more servers and does not require a user interface such as a display, mouse, or the like. In an embodiment, system 10 may be used to implement any of the elements shown in FIG. 1.

System 10 includes a bus 12 or other communication mechanism for communicating information, and a processor 22 coupled to bus 12 for processing information. The processor 22 may be any type of general purpose or special purpose processor. The system 10 also includes a memory 14 for storing instructions and information to be executed by the processor 22. Memory 14 may include any combination of random access memory ("RAM"), read only memory ("ROM"), static storage device such as a magnetic or optical disk, or any other type of computer-readable medium. The system 10 also includes a communication device 20, such as a network interface card, to provide access to a network. Thus, a user may interface with system 10 directly or remotely through a network or any other method.

Computer readable media can be any available media that can be accessed by processor 22 and includes both volatile and nonvolatile media, removable and non-removable media, and communication media. Communication media may include computer readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave or other transport mechanism, and includes any information delivery media.

The processor 22 is also coupled via the bus 12 to a display 24, such as a liquid crystal display ("LCD"). A keyboard 26 and a cursor control device 28, such as a computer mouse, are further coupled to bus 12 to enable a user to interface with system 10.

In one embodiment, memory 14 stores software modules that provide functionality when executed by processor 22. These modules include an operating system 15 that provides operating system functionality for system 10. These modules also include a room demand model module 16 that generates a room demand model to maximize the desired hotel room benefits, as well as all other functions disclosed herein. Since hotel variable operating costs are relatively small, the expected revenue (i.e., the product of the room reservation probability and the room price) is the primary optimization objective in an embodiment. The system 10 may be part of a larger system. Accordingly, system 10 may include one or more additional functional modules 18 to include additional functionality, such as functionality of a property management system ("PMS") (e.g., "Oracle hotel op era property" or "Oracle hotel op era cloud service") or an enterprise resource planning ("ERP") system. A database 17 is coupled to bus 12 to provide centralized storage for modules 16 and 18 and to store guest data, hotel data, transaction data, and the like. In one embodiment, database 17 is a relational database management system ("RDBMS") that may use a structured query language ("SQL") to manage stored data.

In one embodiment, database 17 is implemented as an in-memory database ("IMDB"), particularly when there are a large number of hotel locations, a large number of guests, and a large number of historic data. IMDB is a database management system that relies primarily on main memory for computer data storage. In contrast to database management systems that employ disk storage mechanisms. The main memory database is faster than the disk-optimized database because disk access is slower than memory access, the internal optimization algorithm is simpler and executes fewer CPU instructions. Accessing the data in memory eliminates seek time when querying the data, which provides faster and more predictable performance than a disk.

In one embodiment, database 17, when implemented as an IMDB, is implemented based on a distributed data grid. A distributed data grid is a system in which a collection of computer servers work together in one or more clusters to manage information and related operations, such as computing, within a distributed or clustered environment. The distributed data grid may be used to manage application objects and data that are shared across servers. The distributed data grid provides low response time, high throughput, predictable scalability, continuous availability, and information reliability. In a particular example, a distributed data grid (such as, for example, an "Oracle Coherence" data grid from Oracle corporation) stores information in memory to achieve higher performance and employs redundancy to keep copies of the information synchronized across multiple servers, thereby ensuring the resiliency of the system and ensuring continued availability of data in the event of a server failure.

In one embodiment, system 10 is a computing/data processing system that includes a collection of applications or distributed applications for an enterprise organization, and may also implement logistics, manufacturing, and inventory management functions. The application and computing system 10 may be configured to operate with or be implemented as a cloud-based networking system, a software as a service ("SaaS") architecture, or other type of computing solution.

Embodiments address the problem of predicting demand for multiple hotel room category and service type combinations based on hotel customer attributes, room category and service type characteristics, price offered, and order in which room-rate pairs are presented to the customer. Rather than assuming a homogenous nature of the customer (i.e., where the desired demand should be the same when offering the same price), embodiments assume that the customer base includes several clusters to allow the customer nature and selection pattern to be heterogeneous across clusters. In addition to predicting the needs of these heterogeneous customers (i.e., where the desired needs may be different even when the same price is offered), embodiments also estimate the dynamic size of each cluster and the centroid of each cluster as the calculation is iterated to reflect the new allocation. The main output of the problem is the probability that each individual customer subscribes to a room with a specific room category-service type combination.

Embodiments utilize dynamic clustering methods to enable high accuracy prediction of a reservation customer's room-service combination. Embodiments begin with an initial cluster to divide clients into several clusters such that the characteristics of the clients within each cluster may be more homogenous than the characteristics of the clients from other clusters, and assuming a personalized selection model within each cluster. Since the cluster membership of the clients (i.e., which cluster each client belongs to) is not observable, embodiments employ a soft clustering approach, where the "mix" is captured by the probability that the client belongs to each cluster.

To this end, embodiments implement unsupervised clustering using a random forest clustering algorithm with a number of clusters based on characteristics of potential hotel customers, order of room-service pairs, and their characteristics (including price offered). Next, embodiments derive weighted likelihood functions from observed clients based on a discrete selection polynomial ("MNL") model corresponding to the clusters, where the weights are set to the cluster probabilities obtained from the initial clusters. Then, embodiments maximize the weighted likelihood function to obtain the value of the coefficient for each covariate and the intercept in the MNL model. From these values, a selection probability is calculated for a plurality of hotel room category-service type combinations for each customer. The number of clusters is selected as the value that provides the best prediction accuracy.

In an embodiment, the initial clustering is based on customer characteristics, rather than their selection. To incorporate the customer's selection behavior into the cluster, embodiments update the weights to the initial cluster probability times the selection probability calculated at the previous step, which may be considered as the E-step of the expectation maximization ("EM") algorithm. The embodiment then re-fits the model with the newly formed clusters, performing a dynamic clustering step by maximizing the updated weighted likelihood functions, which constitutes the M-step of the EM algorithm. Finally, the embodiment iterates the E-step and M-step until convergence criteria are met.

After convergence, the embodiment obtains a final estimate of the model parameters. For new customers with their own characteristics, the order of room-service pairs, and room category characteristics (including prices offered), embodiments may predict the probability of selection of new customers by solving the classification problem with a supervised random forest classifier after estimating their relevance to each cluster.

Dynamic iterative reconfigurable clustering algorithm/function

In general, embodiments implement dynamic iterative reconfigurable clustering algorithms/functions for predicting demand to generate hotel room demand models. It is assumed that the customer population of interest consists of a plurality of clusters G, where (G > 1), where patterns within each cluster across the customer's reservation rooms are relatively homogeneous, while heterogeneity exists in the reservation patterns across the clusters. Under this assumption, it is intuitive to consider different G selection models across clusters (i.e., selection models fitted to each cluster separately). However, in practice, the cluster membership indicating which cluster each client belongs to is not observable. In contrast, embodiments implement novel algorithms/functions to address the problem of estimating heterogeneous subscription patterns for clients across clusters when cluster membership is unknown.

In particular, suppose client i is composed of a set of observable covariatesTo characterize, for i=1, …, n, where n is the number of clients in the dataset. Let J be the number of products considered in the market, and S _i For a set of products available to customer i, i.e. < ->Let y _i Representing the product selection made by customer i, where y _i ∈S _i . Product j=1, …, JFrom a set of observable variables->To characterize. The MNL room selection probability within cluster g can then be expressed as follows. Let l _i For the cluster membership indicator of client i,

wherein the method comprises the steps ofIs identifiability and B e {1, …, J } is the baseline product. In the embodiment, b=j is set to demonstrate the embodiment of the present invention. Due to the cluster membership indicator l _i Is not observable and therefore is considered a potential variable and requires a model to interpret the different probabilities belonging to clusters across different customer features. Specifically, let l be called a mixed distribution _i Is shown below:

wherein the method comprises the steps ofIs dependent on->Is unknown.

Pair l _i The general approach to modeling is to assume a MNL (called Logit) model that specifies the probability that a client belongs to cluster g

Wherein the embodiment is provided withIs identifiable. Vector->Specifying how the customer characteristics affect the clusters, i.e. which cluster the customer belongs to. However, with product selection y _i Different, cluster membership indicator l _i Is not observable, so the true structure of the hybrid distribution is unknown in practice and it is difficult to test whether the specified model is correct. The family of parameters of the pre-specified mixture distribution as in equation (1) may not agree with the true mixture distribution, known as a model false assignment problem, which results in biased parameter estimation or low goodness-of-fit measurement, affecting prediction accuracy.

To avoid such model misassignment and improve prediction performance, embodiments achieve half-parametric blending of discrete selection models by assuming equations (1) and (2) instead of equations (1) and MNL model (3). Letting model parameters represent1, …, G }, then ∈>Likelihood functions of (2) are written as

Wherein for the followingNo pre-specified parametric model form is imposed, which can be estimated by using any non-parametric clustering method, such as random forests. In other embodiments, other unsupervised machine learning techniques for clustering may also be used. Then, the embodiment uses the same ideas as for the EM algorithm as follows. Assume potential cluster membership indicator l _i Is known to be. Then the complete likelihood function is

And the complete log-likelihood function is

In the EM algorithm, the maximizer of the objective function of equation (4) can be found by using the following iterative method:

specifically, the following E-step and M-step are repeated as follows.

E-step

Given observed dataCalculation of l _i Conditional expectation of (2)

M-step

Updating parameters by solving the following equation

As disclosed, embodiments employ unsupervised clustering techniques, such as random forests, based on customer characteristics. Thus, the EM algorithm may be adapted to the context, referred to as iterative reconfigurable clustering, as follows.

Initial clustering:

learning unsupervised soft clustering (e.g., random forest, k-means) is performed at G. Clustering is based on client-level covariatesAnd obtaining the cluster probability of each cluster +.>Make->Wherein->Is->Is used to estimate the initial estimate of (a). For->We find the initial parameter values by solving the following equation:

e-step

Embodiments determine conditional cluster probabilities by using discrete selection models of observed selections and fits, as follows: if y _i J, then

So that

M-step

Updating selected model parameters to be by solving the following equationAnd->For each g, first find +.>For j=1, …, the solution for J-1 is shown below.

Then through aboutSolving the following equation to obtain->

Where g=2, …, G.

Repeating (E-step) and (M-step) until, for any ε > 0, the conditions asIs a convergence criterion of (1).

The above may be regarded as a variant of the EM algorithm. In an embodiment, the theorem of "Dempster et al (1977)" may be applied to the proposed iterative algorithm, which indicates the solutionAstringe to->Wherein->Is our objective functionIs the maximization of (2).

Prediction of room category and rate code combinations

After convergence, the embodiment obtains model parametersIs used to estimate the final estimate of (a). For by->The new customer characterized, for example, predicts the selection probability as follows: let S ^* For a usable product, for j ε S ^* ，

Wherein the method comprises the steps ofIs the probability of belonging to cluster g predicted by soft clustering, and +.>Is the feature vector of room j available to the new customer.

FIG. 3 is a flow diagram of the functionality of the room demand model module 16 of FIG. 2 for generating a room demand model, according to one embodiment. In one embodiment, the functions of the flow chart of fig. 3 are implemented by software stored in a memory or other computer-readable or tangible medium and executed by a processor. In other embodiments, the functionality may be performed by hardware (e.g., through the use of application specific integrated circuits ("ASICs"), programmable gate arrays ("PGAs"), field programmable gate arrays ("FPGAs"), etc.), or any combination of hardware and software.

At 302, an initial unsupervised soft cluster is developed to cluster clients based on a plurality of attributes/characteristics assigned to each client. In an embodiment, the attributes may include one or more of the following: (1) Global distribution systems (e.g., amadeus, SABRE, etc.) used; (2) booking channels; (3) number of nights; (4) number of arrival clients; (5) subscribing to the number of days in advance; (6) weekends and weekdays; (7) enterprise booking.

The initial clustering at 302 is based on customer characteristics, not customer selections. Customer characteristics are those known at the time of the request for a room and include data such as arrival date and time, the number of people in the community, and the reservation channel. In addition, the customer characteristic data includes other inferred characteristics, such as a reservation window (i.e., time between reservation and arrival date).

Both the initial clustering and the dynamic clustering described below (where the initial and subsequent clusters are dynamically updated) incorporate machine learning. Specifically, the initial clustering at 302 may incorporate any unsupervised machine learning technique for clustering, such as random forests or soft clustering algorithms using a gaussian mixture model. Unlike the customer's selection, cluster membership is not observable, so pre-specified parametric models on how to form clusters are more challenging based on customer characteristic assumptions, and it is difficult to test if the pre-specified parametric models are correct. Failure to specify the correct model results in biased parameter estimates or low goodness-of-fit measurements, which affect prediction accuracy. Since embodiments do not require any pre-specified parametric model form for the cluster structure, possible bias from model false assignments can be avoided.

FIG. 4 illustrates an example of initial clustering according to an embodiment. As shown in fig. 4, three clusters are formed based on guest characteristics, external factors, and travel attributes. In an embodiment, the number of clusters is a predefined parameter based on the interpretability of the clusters, which typically limits the number of clusters to a single digit. In various embodiments, two to four clusters are used.

At 304, embodiments estimate an initial mix of multiple Logit ("MNL") models of demand for hotel room category and rate code combinations based on parameters related to hotel room supply, including: (1) a price offered; (2) room category and rate plan location in the offer; and (3) room and rate characteristics such as field of view, room size, whether breakfast is contained, free cancellation, etc. For each cluster formed at 302, a separate MNL model is built at 304. Fig. 5 is an example illustrating various offered prices (e.g., $335), room categories (e.g., luxury or premium, extra large or large) and rate codes (e.g., "rates containing breakfast"). At 304, historical subscription data is stored in a database (e.g., database 17 of FIG. 2) to estimate the values defined in connection with equation (4) above by solving equation (5) aboveParameters.

FIG. 6 illustrates selection modeling of guest clusters according to an example embodiment. As shown in fig. 6, each cluster uses a unique discrete selection model to predict each customer's selection of hotel room and rate code combination. Fig. 7 illustrates initial assignment of MNL models to each cluster according to an embodiment.

306. 308 and 310 together and on an iterative basis form a desired maximization ("EM") function. The EM function includes 306, 308, and 310, where it also contains soft clusters updated in the E-step at 306. The soft cluster at 302 is an initial cluster that is not repeated. At 306, for the desired "E-step", the cluster probability is updated by combining the selection probabilities of the clients evaluated at the parameter values of the current iteration.

Fig. 8 illustrates a proposed likelihood function for use with EM functions according to an embodiment. As shown, the proposed likelihood functions include both the cluster model generated at 302 and shown in fig. 4, and the selection model generated at 304 and shown in fig. 6. The proposed likelihood function is the objective function of the EM function. Embodiments find the maximizer of the objective function by using the EM function to estimate the model parameters.

Fig. 9 illustrates a portion of EM functionality according to an embodiment. After the "initial step" (i.e., the soft clustering performed at 302), the desired E-step is determined at 306.

At 308, for maximizing the "M-step," the embodiment estimates an updated mix of the MNL model, where the mix probability is the updated cluster probability in the E-step. At 310, 306 and 308 are repeated until convergence criteria are met: new prediction error-old prediction error <0.0001.

At 312, a demand model is generated that predicts the probability of selection of the room category and rate code combination for the new customer using the estimated parameters from 306, 308. At 314, the function ends.

The functionality of fig. 3 combines the estimation of discrete selection modeling with data-driven customer segment identification and captures the different preferences of heterogeneous customer populations and provides interpretable model output. The demand model generated at 312 provides a practical way that can help hotel operators parse their profiles based on their customer/guest preferences, which can be used as a valuable input to: (1) Formulating a more efficient marketing strategy and providing personalized recommendations that are more likely to be accepted; and (2) generating an optimal personalized price and display location for each room type (e.g., suite with waterscape and big bed).

Fig. 10 illustrates a portion of EM functionality according to an embodiment. Fig. 10 illustrates the M-step at 308 and the repetition up to convergence at 310.

Fig. 11-16 illustrate examples of embodiments of the invention for three clusters. Fig. 11 illustrates soft clustering at 1101 (302 of fig. 3) and selection modeling at 1102 (304 of fig. 3), wherein a different MNL model is generated for each cluster according to the soft clustering. The number of clusters is predetermined before using the EM function. To select the optimal number of clusters, the prediction accuracy measurements are compared across several different numbers of clusters, and the optimal number that achieves the most accurate prediction is selected. There is a different MNL for each cluster, but all model parameters are jointly estimated. The initial data 1103 for each guest is used as input and an initial cluster probability 1104 is generated for each guest.

Fig. 12 illustrates a first iteration (306 and 308 of fig. 3) of reassigning conditional cluster probabilities using an E-step at 1201.

Fig. 13 illustrates a first iteration (306 and 308 of fig. 3) of updating the selection model using an M-step at 1301, which corrects the conditional cluster probability.

Fig. 14 illustrates a second iteration using the E-step updated conditional cluster probability at 1301, and fig. 15 illustrates a second iteration using the M-step. For purposes of illustration, assume that there are only two iterations.

Then, FIG. 16 illustrates the generation of a demand model using estimated model parameters, which forms a prediction of the selection probability of the new customer.

Metrics for evaluation

To study the performance of iterative reconfigurable clusters according to embodiments, embodiments divide the dataset into a training dataset and a test dataset. After estimating model parameters from the training data and initial clustering, embodiments obtain predictions of product selection among customers in the test data. For prediction accuracy measurements, embodiments use the correct classification rate ("CCR") and mean square error ("MSE").

CCR is calculated as the percentage of observations where the option with the highest predictive probability coincides with the observed selection.

The MSE is calculated as:

wherein y is _i Andfor customer i to select true and predicted room type j, φ _te Indexing for clients in test dataSet, and n _te For the number of clients in the test data. This metric, also known as the Brier score, is commonly used to evaluate probabilistic predictions.

In experiments, embodiments were applied using real hotel datasets by utilizing proprietary datasets from multiple hotels in multiple cities and countries. These data include subscription information and corresponding customer characteristics. In addition, log data (i.e., real-time client requests for subscriptions and corresponding responses to the subscription server system) are included. Information is extracted from the log regarding the order of display of the room and rate codes. The order is strategically entered by each hotel operator so that each customer has a different display order. The final dataset contained 9,173 subscriptions from 2019, 7, 2, to 2019, 7, 19, with 18 different rooms and 15 different rate codes.

The embodiment first finds the number of optimal clusters in the customer population, which is typically unknown. Embodiments employ a prediction criterion method to select the number of optimal clusters. In particular, embodiments take MSE and select the number of clusters of 2, 3, 4, and 5 that have the highest prediction accuracy. Experiments determined that 2 clusters had the best performance among the 4 options.

Given 2 clusters, embodiments use a real hotel reservation dataset to implement embodiments of the invention. Specifically, the prediction accuracy of the embodiments is compared to a single cluster reference (benchmark). The example divides each dataset into a training dataset (80%) and a test dataset (20%). The results are presented in table 1 below, with the iteration stopped at 17, because the MSE-based criterion (i.e., near 0.0001) is satisfied:

TABLE 1 comparison of prediction accuracy with iteration using MSE and CCR

FIG. 17 illustrates a comparison of prediction accuracy with iteration between CCR and MSE according to embodiments of the present invention. In fig. 11, lines 1701 and 1702 are for 2 clusters and lines 1703 and 1704 are for a single cluster. As shown, the predictive performance improves with iteration for 2 clusters using embodiments of the invention. Specifically, the value of CCR is becoming increasing, while the value of MSE is becoming decreasing with iteration. It was also observed that the results of a single cluster (i.e. the known solution) were worse than the results of 2 clusters.

Fig. 18 illustrates how the cluster characteristics change with iteration given two clusters, according to an embodiment. Specifically, fig. 18 illustrates how the centroid value moves with iteration, with curve 1850 for CCR and curve 1860 for MSE. Clients are clustered using seven attributes to solve the problem of heterogeneous client groups: a global distribution system (1803), a reservation channel (1802), a night number (1806), a number of arriving customers (1807), a number of days ahead reserved (1801), whether the customers arrived on the weekend (1805), and whether the customers were reserved by an enterprise code (1804). Fig. 18 shows how each attribute moves with iteration for each cluster.

As disclosed, embodiments incorporate a novel method to predict customer selection and estimate relative values of room category and service type characteristics in the hotel industry based on the attributes of the subscribing customers, the order of room-service pairs in the offers, and the prices offered. In particular, most demand prediction tools currently in use by the hotel industry aim to provide a total number of reservations based on a time series analysis assuming a single cluster (i.e., a homogenous customer group), thereby ignoring heterogeneous customer groups. These demand modeling tools tend to be ineffective when there are heterogeneous customers with significantly different willingness-to-pay and behavioral patterns. Even though some tools consider heterogeneous customer groups, they employ standard clustering algorithms that may not reflect customer selection behavior during the clustering process. Furthermore, in general, no demand forecast tool addresses the order of room category-rate code pairs. In addition to the price offered, the order of display on the website also affects the customer's selection behavior.

Embodiments enable high accuracy prediction of a reservation customer's room-service combination. Through calculation experiments, the embodiment shows that the prediction rate using the dynamic clustering method is about 4% higher than that of the static clustering method. Further, embodiments input information regarding the order of room categories and rate codes for the display optimization system, which may help hotel operators formulate more appropriate marketing strategies and present personalized recommendations that are more prone to acceptance.

Furthermore, embodiments may incorporate any unsupervised machine learning technique for clustering (such as random forests or soft clustering algorithms using gaussian mixture models) into the first step of the algorithm. Unlike the customer's selection, cluster membership is not observable, and therefore, pre-specified parametric models on how to form clusters are more challenging based on customer characteristic assumptions, and it is difficult to test whether the pre-specified parametric models are correct. Failure to specify the correct model results in biased parameter estimates or low goodness-of-fit measurements, which affect prediction accuracy. Since embodiments do not require any pre-specified parametric model form for the cluster structure, possible bias from model false assignments can be avoided.

Embodiments implement dynamic clustering as a form of machine learning, particularly when it involves training as in embodiments of the present invention. Embodiments use unsupervised learning that takes a dataset containing only inputs and finds structures in the data, such as groupings or clusters of data points. Cluster analysis is the assignment of a set of observations into subsets (called clusters) such that observations within the same cluster are similar and observations from different clusters are dissimilar according to one or more pre-specified criteria. Different clustering techniques make different assumptions about the structure of the data, which are often defined by some measure of similarity and evaluated, for example, by internal closeness or similarity between members of the same cluster, and degree of separation (differences between clusters). Dynamic clustering as a form of unsupervised online/incremental machine learning considers two concepts: (1) an increment of a learning method for designing a cluster model; and (2) adaptation of the learned model (parameters and structure).

The features, structures, or characteristics of the disclosure may be combined in any suitable manner in one or more embodiments. For example, use of the terms "one embodiment," "some embodiments," "certain embodiments," or other like language throughout this specification refers to the fact that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the present disclosure. Thus, appearances of the phrases "one embodiment," "some embodiments," "certain embodiments," or other similar language throughout this specification do not necessarily all refer to the same group of embodiments, and the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

Those of ordinary skill in the art will readily appreciate that the embodiments described above may be practiced with steps in a different order and/or with elements of a different configuration than those disclosed. Thus, while the present disclosure contemplates the summarized embodiments, it will be apparent to those skilled in the art that certain modifications, variations, and alternative constructions will be apparent, while remaining within the spirit and scope of the disclosure. Accordingly, to determine the boundaries and bounds of the disclosure, reference should be made to the appended claims.

Claims (20)

1. A method of generating a demand model for potential hotel clients of a hotel room, the method comprising:

forming a plurality of clusters based on characteristics of potential hotel clients, each cluster including a corresponding weight and cluster probability;

generating an initial estimated mixture of multiple logic (MNL) models corresponding to each of the plurality of clusters, the mixture of MNL models including a weighted likelihood function based on features and weights;

determining the corrected clustering probability and updating the weight;

estimating an updated estimated MNL model based on the modified cluster probability and the updated weights and maximizing a weighted likelihood function; and

a demand model is generated that is adapted to predict a selection probability of a combination of room categories and rate codes for potential hotel customers based on the updated estimated mix of updated weights and MNL models.

2. The method of claim 1, wherein characteristics of the potential hotel customer are known when the potential customer requests a hotel room.

3. The method of claim 1, generating an estimated mix of MNL models based on the provided prices, room categories and rate plan locations in the quotes, and room and rate characteristics.

4. The method of claim 1, wherein forming the plurality of clusters comprises unsupervised machine learning.

5. The method of claim 4, wherein the unsupervised machine learning comprises one of dynamic clustering or soft clustering using a gaussian mixture model.

6. The method of claim 1, wherein the determining and estimating are repeated until convergence criteria are reached, and the demand model is generated after the convergence criteria are reached.

7. The method of claim 1, wherein the features comprise at least one of: arrival date and time, number of people in the community, reservation channel or reservation window.

8. The method of claim 1, wherein the demand model is adapted to maximize revenue for hotel rooms.

9. A computer-readable medium having instructions stored thereon that, when executed by one or more processors, cause the processors to generate a demand model for potential hotel customers of a hotel room, the generating the demand model comprising:

forming a plurality of clusters based on characteristics of potential hotel clients, each cluster including a corresponding weight and cluster probability;

generating an initial estimated mixture of multiple logic (MNL) models corresponding to each of the plurality of clusters, the mixture of MNL models including a weighted likelihood function based on features and weights;

determining the corrected clustering probability and updating the weight;

estimating an updated estimated mix of the MNL model based on the modified cluster probabilities and the updated weights and maximizing a weighted likelihood function; and

a demand model is generated that is adapted to predict a selection probability of a combination of room categories and rate codes for potential hotel customers based on the updated estimated mix of updated weights and MNL models.

10. The computer readable medium of claim 9, wherein characteristics of the potential hotel client are known when the potential client requests a hotel room.

11. The computer readable medium of claim 9, generating an estimated mix of MNL models is based on the provided prices, room categories and rate plan locations in the quote, and room and rate characteristics.

12. The computer-readable medium of claim 9, wherein forming the plurality of clusters comprises unsupervised machine learning.

13. The computer-readable medium of claim 12, wherein the unsupervised machine learning comprises one of dynamic clustering or soft clustering using a gaussian mixture model.

14. The computer-readable medium of claim 9, wherein the determining and estimating are repeated until a convergence criterion is reached, and the demand model is generated after the convergence criterion is reached.

15. The computer-readable medium of claim 9, wherein the features comprise at least one of: arrival date and time, number of people in the community, reservation channel or reservation window.

16. The computer readable medium of claim 9, wherein the demand model is adapted to maximize revenue for a hotel room.

17. A hotel reservation system for generating a demand model for potential hotel clients of a hotel room, comprising:

one or more processors coupled with the stored instructions; and

a database storing historical subscription data;

the processor is configured to:

forming a plurality of clusters based on characteristics of potential hotel clients, each cluster including a corresponding weight and cluster probability;

generating an initial estimated mixture of multiple logic (MNL) models corresponding to each of the plurality of clusters, the mixture of MNL models including a weighted likelihood function based on features and weights;

determining the corrected clustering probability and updating the weight;

estimating an updated estimated mix of the MNL model based on the modified cluster probabilities and the updated weights and maximizing a weighted likelihood function; and

a demand model is generated that is adapted to predict a selection probability of a combination of room categories and rate codes for potential hotel customers based on the updated estimated mix of updated weights and MNL models.

18. The hotel reservation system of claim 17, wherein characteristics of the potential hotel client are known when the potential client requests a hotel room.

19. The hotel reservation system of claim 17, generating an estimated MNL model based on the provided prices, room categories and rate plan locations in the quotes, and room and rate characteristics.

20. The hotel reservation system of claim 17, wherein forming the plurality of clusters comprises unsupervised machine learning.