Hotel price monitoring method and system based on multiple data sources and machine learning
By employing a multi-data source and machine learning approach to hotel price monitoring, and utilizing web crawling technology and machine learning algorithms to build a price prediction model, this method solves the problems of low efficiency and poor accuracy in traditional monitoring methods. It enables real-time monitoring and early warning of hotel prices, thus protecting consumer rights.
Patent Information
- Application Number
- CN202511048575.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-29
- Publication Date
- 2025-11-21
AI Technical Summary
In existing technologies, hotel price monitoring relies on manual inspections and single data analysis, which suffers from low monitoring efficiency, limited scope, inability to detect abnormal price fluctuations in a timely manner, and difficulty in comprehensively and accurately determining whether hotel prices are abnormal.
A hotel price monitoring method based on multiple data sources and machine learning is adopted. Data is collected from multiple platforms and databases through web crawling technology, preprocessed, and fused to build a price prediction model of random forest and gradient boosting decision tree, so as to realize real-time monitoring and early warning.
It improves the accuracy and timeliness of monitoring, reduces the cost of manual intervention, enables timely detection of price anomalies, maintains price order in the tourism market, and protects consumer rights.
Smart Images
Figure CN120996989A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of big data processing, in particular to a hotel price monitoring method and system based on multiple data sources and machine learning. BACKGROUND
[0002] During holidays, the demand for tourism travel of residents increases significantly, and hotels, as an important supporting service facility for tourism travel, have frequent price fluctuations. However, some hotels take advantage of the holiday tourism peak season to raise prices, resulting in "sky-high hotels", which seriously damages the rights and interests of consumers and disrupts the order of the tourism market. At present, the traditional hotel price monitoring method mainly relies on manual patrol and sampling analysis of a small amount of data, which has the problems of low monitoring efficiency, limited monitoring range, and inability to timely discover abnormal price fluctuations. Although there are some monitoring methods based on web crawler technology to obtain hotel price data, it is difficult to comprehensively and accurately determine whether the hotel price has abnormal fluctuations by relying on single price data, which cannot meet the actual monitoring needs.
[0003] How to integrate multiple data sources and use advanced algorithms to monitor hotel prices is a technical problem to be solved. SUMMARY
[0004] The technical task of the present application is to solve the above problems, and provide a hotel price monitoring method and system based on multiple data sources and machine learning, to solve the technical problem of how to integrate multiple data sources and use advanced algorithms to monitor hotel prices.
[0005] In a first aspect, the present application provides a hotel price monitoring system based on multiple data sources and machine learning, which comprises a data acquisition module, a data preprocessing module, a data fusion module, a model training module and a monitoring and early warning module.
[0006] The data acquisition module is used to acquire multiple source data from multiple tourism platforms and tourism management department databases based on web crawler technology.
[0007] The data preprocessing module is used to preprocess the acquired multiple source data, and through preprocessing, cleaning operation, conversion operation and normalization operation, the preprocessed multiple source data is obtained.
[0008] The data fusion module is used to fuse the preprocessed multiple source data to form a feature vector comprising hotel price related information.
[0009] The model training module is used to construct a price prediction model based on machine learning method, and to train and test the price prediction model based on the feature vector, to obtain the trained price prediction model.
[0010] The monitoring and early warning module is configured to call the data collection module to collect real-time multi-source data, call the data preprocessing module to preprocess the real-time multi-source data to obtain preprocessed multi-source data, call the data fusion module to fuse the preprocessed multi-source data to form a feature vector, and input the feature vector into the trained price monitoring model to determine whether the hotel price has an abnormal fluctuation, predict the output abnormality degree type and the probability value corresponding to the abnormality degree type as a prediction result, and send an early warning message based on the prediction result.
[0011] Preferably, the multi-source data includes hotel name, hotel geographical location, room type information, price, occupancy rate, user evaluation, holiday tourist flow prediction data, and local hotel industry average price, wherein the price includes current day price, historical price, and holiday price.
[0012] When collecting multi-source data based on network crawler technology from multiple tourism platforms and tourism management department databases, the network crawler sets access frequency and request header information according to needs.
[0013] Preferably, repeated data, data with too many missing values, and obviously erroneous data are removed through cleaning operations, different formats of data are converted to a unified predetermined format through conversion operations, data of different magnitudes are standardized through normalization operations, and text data is processed through natural language processing, which includes word segmentation and stop word removal.
[0014] Preferably, the model training module is configured to perform the following operations:
[0015] Data set division: the feature vector is divided into a training set and a test set through stratified sampling;
[0016] Feature engineering: for continuous data in the training set and the test set, the data is standardized to obtain processed features; for discrete data in the training set and the test set, the data is one-hot encoded to obtain processed features; for textual data including user evaluations, the data is converted into numerical features to obtain processed features, and based on the feature importance ranking of the random forest, features that have an impact on price anomalies are selected from the processed features to obtain sample features;
[0017] Model training: a price prediction model is constructed based on a machine learning method, the price prediction model is trained and tested based on the sample features, and a trained price prediction model is obtained;
[0018] Preferably, the price prediction model is a network model constructed based on a random forest and a gradient boosting decision tree, and the model training of the price prediction model includes the following:
[0019] The random forest model is trained.
[0020] Model training on the gradient boosting decision tree model;
[0021] Integrated model construction: adjust the weight through cross-validation to make the integrated model optimal in performance on the test set, integrate the prediction results of the two models by weighted voting, and weighted average according to the probability value output by the model;
[0022] The training steps of the random forest model are as follows:
[0023] Bagging sampling: multiple subsets are extracted from the training set with replacement, and a decision tree is trained for each subset;
[0024] Feature random selection: when the node of each tree is split, a part of features is randomly selected to calculate the best split point;
[0025] The training steps of the gradient boosting decision tree model are as follows:
[0026] Initialization of prediction: use the mean value of all samples as the initial prediction value;
[0027] Calculate the residual: the difference between the current model prediction value and the true value is used as the target of the next tree;
[0028] Iterative training: each tree fits the residual of the previous round, and minimizes the loss function through gradient descent;
[0029] Regularization: control overfitting through learning_rate and subsample.
[0030] As preferred, the early warning information further includes an abnormal hotel name, an abnormal time, an abnormal price, and an abnormal reason analysis, and the abnormal reason analysis is generated based on the prediction result;
[0031] The monitoring and early warning module is used to push the early warning information in the form of short message, email and APP push.
[0032] In a second aspect, the present application provides a hotel price monitoring method based on multiple data sources and machine learning, comprising the following steps:
[0033] Data acquisition: based on the network crawler technology, multiple source data are collected from multiple tourism platforms and tourism management department databases;
[0034] Data preprocessing: the collected multiple source data are preprocessed, and through preprocessing, cleaning operation, conversion operation and normalization operation are performed to obtain preprocessed multiple source data;
[0035] Data fusion: the preprocessed multiple source data are fused to form a feature vector including hotel price related information;
[0036] Model training: a price prediction model is constructed based on a machine learning method, and model training and testing are performed on the price prediction model based on the feature vector to obtain the trained price prediction model;
[0037] Monitoring and early warning: real-time multi-source data is collected, preprocessed, and fused to form a feature vector, and the trained price monitoring model is used to determine whether the hotel price has abnormal fluctuations, predict the output abnormality degree type and the probability value corresponding to the abnormality degree type as the prediction result, and send early warning information based on the prediction result.
[0038] As a preferred embodiment, the multi-source data includes hotel name, hotel geographic location, room type information, price, occupancy rate, user evaluation, holiday tourist flow prediction data, and local hotel industry average price, wherein the price includes current day price, historical price, and holiday price.
[0039] When collecting multi-source data from multiple tourism platforms and tourism management department databases based on web crawler technology, the web crawler sets the access frequency and request header information according to the needs.
[0040] As a preferred embodiment, the cleaning operation removes duplicate data, data with too many missing values, and obviously incorrect data, the conversion operation converts data of different formats into a unified predetermined format, the normalization operation standardizes data of different magnitudes, and natural language processing is performed on text data, including word segmentation and stop word removal.
[0041] As a preferred embodiment, model training includes the following operations:
[0042] Data set division: the feature vector is divided into a training set and a test set by stratified sampling;
[0043] Feature engineering: for continuous data in the training set and the test set, the data is standardized to obtain processed features; for discrete data in the training set and the test set, the data is one-hot encoded to obtain processed features; for textual data including user evaluations, the data is converted into numerical features to obtain processed features, and based on the feature importance ranking of the random forest, features that have an impact on price anomalies are selected from the processed features to obtain sample features;
[0044] Model training: a price prediction model is constructed based on a machine learning method, and model training and testing are performed on the price prediction model based on the sample features to obtain the trained price prediction model;
[0045] Wherein, the price prediction model is a network model constructed based on random forest and gradient boosting decision tree, and model training of the price prediction model comprises the following:
[0046] Model training is performed on the random forest model;
[0047] Model training is performed on the gradient boosting decision tree model;
[0048] Integrated model construction: the performance of the integrated model on the test set is optimized by adjusting the weight through cross-validation, the prediction results of the two models are integrated by using the weighted voting method, and the weighted average is obtained according to the probability value output by the model;
[0049] The training steps of the random forest model are as follows:
[0050] Bagging sampling: multiple subsets are extracted from the training set with replacement, and a decision tree is trained for each subset;
[0051] Feature random selection: when the node of each tree is split, a part of features is randomly selected to calculate the best split point;
[0052] The training steps of the gradient boosting decision tree model are as follows:
[0053] Initialization prediction: the mean value of all samples is used as the initial prediction value;
[0054] Residual error calculation: the difference between the current model prediction value and the true value is used as the target of the next tree;
[0055] Iterative training: each tree fits the residual error of the previous round, and the loss function is minimized through gradient descent;
[0056] Regularization: control overfitting through learning_rate and subsample.
[0057] The hotel price monitoring method and system based on multiple data sources and machine learning have the following advantages:
[0058] 1. Improve monitoring accuracy: collect multi-dimensional data, compared with the traditional single price data monitoring method, can more comprehensively reflect the influencing factors of hotel prices, improve the accuracy of monitoring, for example, combined with user evaluation can understand the influence of hotel service quality on price, through holiday tourist flow prediction data can predict the price fluctuation trend in advance;
[0059] 2. Real-time monitoring and timely warning: real-time dynamic monitoring and timely warning of hotel prices are realized, so that the management department can quickly take measures to intervene and maintain the price order of the tourism market, and also provides timely information for consumers to protect their legitimate rights and interests;
[0060] 3. The cost of human intervention is reduced: the application of multi-source data fusion and machine learning algorithms improves monitoring efficiency, reduces human intervention, reduces monitoring costs, and has good economic and social benefits. BRIEF DESCRIPTION OF DRAWINGS
[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0062] The present application will be further described below in conjunction with the drawings.
[0063] Figure 1 A flowchart of a hotel price monitoring method based on multiple data sources and machine learning for embodiment 1. DETAILED DESCRIPTION
[0064] The present application will be further described below in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it. However, the embodiments are not limiting the present application, and the technical features in the embodiments and the embodiments can be combined with each other without conflict.
[0065] The present application provides a hotel price monitoring method and system based on multiple data sources and machine learning, which is used to solve the technical problem of how to integrate multiple data sources and use advanced algorithms to realize hotel price monitoring.
[0066] Embodiment 1:
[0067] The present application provides a hotel price monitoring method and system based on multiple data sources and machine learning, which is used to solve the technical problem of how to integrate multiple data sources and use advanced algorithms to realize hotel price monitoring.
[0068] The data acquisition module is used to acquire multi-source data from multiple tourism platforms and tourism management department databases based on web crawler technology.
[0069] In this embodiment, the multi-source data includes hotel name, hotel geographic location, room type information, price, occupancy rate, user evaluation, holiday tourist flow prediction data, and local hotel industry average price, wherein the price includes current price, historical price and holiday price. When multi-source data is acquired from multiple tourism platforms and tourism management department databases based on web crawler technology, the web crawler sets the access frequency and request header information according to the needs.
[0070] As a specific implementation of the data collection module, the module uses web crawler technology to collect data from multiple travel booking platforms (such as Qunar, Meituan, Feizhu, etc.), hotel official websites, social media platforms (such as Weibo, Xiaohongshu, etc.), and local tourism management department databases. The collected data includes but is not limited to hotel name, geographic location, room type information, price data (current price, historical price, holiday price), occupancy rate, user reviews, holiday tourist flow prediction data, local hotel industry average price data, and other multi-source data.
[0071] To ensure the legality and stability of data collection, the web crawler sets a reasonable access frequency and request header information to avoid excessive pressure on the target website and violation of relevant regulations.
[0072] The data preprocessing module is used to preprocess the collected multi-source data, through preprocessing, cleaning, conversion, and normalization operations, to obtain preprocessed multi-source data.
[0073] Among them, through the cleaning operation to remove duplicate data, data with too many missing values and obviously incorrect data, through the conversion operation to convert data of different formats into a unified predetermined format, through the normalization operation to standardize data of different magnitudes, and to process natural language processing for text data, natural language processing includes word segmentation and stop word removal.
[0074] As a specific implementation of the data preprocessing module, the module preprocesses the collected multi-source data by cleaning, converting, and normalizing. Cleaning data removes duplicate data, data with too many missing values, and obviously incorrect data; converting data unifies different formats of data into a format that the system can recognize and process; normalization processing standardizes data of different magnitudes so that data is in the same range, facilitating subsequent analysis. At the same time, text data (such as user reviews) is subjected to natural language processing operations such as word segmentation and stop word removal to extract key information.
[0075] The data fusion module is used to fuse the preprocessed multi-source data to form a feature vector including hotel price related information.
[0076] As a specific implementation of the data fusion module, the module fuses the preprocessed data to construct a multi-dimensional hotel price monitoring dataset. By using feature fusion, data features of different sources and types are integrated to form a feature vector containing multiple aspects of hotel price related information, providing comprehensive data support for subsequent analysis.
[0077] The model training module is used to construct a price prediction model based on machine learning methods, and to train and test the price prediction model based on the feature vector to obtain the trained price prediction model.
[0078] In this embodiment, the model training module is configured to perform the following operations:
[0079] (1) Data set division: the feature vectors are divided into a training set and a test set by stratified sampling;
[0080] (2) Feature engineering: for continuous data in the training set and the test set, the data is standardized to obtain processed features; for discrete data in the training set and the test set, the data is one-hot encoded to obtain processed features; for textual data including user reviews, the data is converted into numerical features to obtain processed features, and based on the feature importance ranking of the random forest, features that have an impact on price anomalies are selected from the processed features to obtain sample features;
[0081] (3) Model training: a price prediction model is constructed based on a machine learning method, the price prediction model is trained and tested based on the sample features, and a trained price prediction model is obtained.
[0082] The price prediction model is a network model constructed based on a random forest and a gradient boosting decision tree, and the model training of the price prediction model includes the following:
[0083] (1) Model training of the random forest model;
[0084] (2) Model training of the gradient boosting decision tree model;
[0085] (3) Ensemble model construction: the weights are adjusted through cross-validation to make the ensemble model optimal in performance on the test set, the prediction results of the two models are integrated by weighted voting, and the weighted average is calculated according to the probability value output by the model.
[0086] The training steps of the random forest model are as follows:
[0087] (1) Bagging sampling: multiple subsets are extracted from the training set with replacement, and each subset is used to train a decision tree;
[0088] (2) Random feature selection: when each tree node is split, a part of features is randomly selected to calculate the best split point.
[0089] The training steps of the gradient boosting decision tree model are as follows:
[0090] (1) Initialization of prediction: the mean value of all samples is used as the initial prediction value;
[0091] (2) Calculation of residual error: the difference between the current model prediction value and the true value is used as the target of the next tree;
[0092] (3) Iterative training: each tree fits the residual of the previous round and minimizes the loss function through gradient descent;
[0093] (4) Regularization: control overfitting through learning_rate and subsample.
[0094] The monitoring and early warning module is configured to call the data acquisition module to acquire real-time multi-source data, call the data preprocessing module to preprocess the real-time multi-source data to obtain preprocessed multi-source data, call the data fusion module to fuse the preprocessed multi-source data to form a feature vector, and input the feature vector to determine whether the hotel price has an abnormal fluctuation through the trained price monitoring model, predict an abnormal degree type and a probability value corresponding to the abnormal degree type as a prediction result, and send an early warning information based on the prediction result.
[0095] In this embodiment, the early warning information further includes an abnormal hotel name, an abnormal time, an abnormal price, and an abnormal reason analysis, and the abnormal reason analysis is generated based on the prediction result.
[0096] The monitoring and early warning module is configured to push the early warning information in the form of a short message, an email, and an APP push.
[0097] The module inputs the data acquired in real time and preprocessed and fused into a trained machine learning model, the model analyzes and predicts the input data to determine whether the hotel price has an abnormal fluctuation. When the model determines that the hotel price has an abnormal fluctuation, the system sends early warning information to relevant management departments and consumers in various ways such as short messages, emails, and APP pushes according to the abnormal degree. At the same time, the system generates a detailed abnormal fluctuation report including an abnormal hotel name, an abnormal time, an abnormal price, an abnormal reason analysis (based on multi-source data feature analysis), and other contents, which provides a basis for management departments to supervise and consumers to make decisions.
[0098] The system of this embodiment uses web crawler technology to collect data from multiple tourism booking platforms, hotel official websites, social media platforms, and local tourism management department databases; uses an ensemble learning algorithm combined with a random forest algorithm and a gradient boosting decision tree algorithm to build a hotel price abnormal fluctuation prediction model; inputs real-time collected and preprocessed and fused data into a trained machine learning model to determine whether the hotel price has an abnormal fluctuation, and sends early warning information to relevant management departments and consumers according to the abnormal degree. The system realizes intelligent monitoring and early warning of tourism holiday hotel price abnormal fluctuations by fusing multi-source data and using machine learning algorithms for analysis and processing, improves the accuracy and timeliness of monitoring, and maintains the price order of the tourism market.
[0099] Embodiment 2:
[0100] The application discloses a hotel price monitoring method based on multiple data sources and machine learning, which comprises five steps of data collection, data preprocessing, data fusion, model training and monitoring and early warning.
[0101] Step S100 data collection: based on the network crawler technology, multiple source data are collected from multiple tourism platforms and tourism management department databases.
[0102] In the embodiment, the multiple source data comprise hotel name, hotel geographical position, room type information, price, occupancy rate, user evaluation, holiday tourism passenger flow prediction data and local hotel industry average price, wherein the price comprises current day price, historical price and holiday price. When the multiple source data are collected based on the network crawler technology from the multiple tourism platforms and the tourism management department databases, the network crawler sets access frequency and request header information according to needs.
[0103] As a specific implementation of data collection, the network crawler technology is used to collect data from multiple tourism booking platforms (such as Ctrip, Meituan, Feizhu and the like), hotel official websites, social media platforms (such as Weibo, Xiaohongshu and the like) and local tourism management department databases. The collected data comprise but are not limited to hotel name, geographical position, room type information, price data (current day price, historical price and holiday price), occupancy rate, user evaluation, holiday tourism passenger flow prediction data, local hotel industry average price data and the like.
[0104] In order to ensure the legality and stability of data collection, the network crawler sets reasonable access frequency and request header information, so as to avoid excessive pressure on the target website and violation of relevant regulations.
[0105] Step S200 data preprocessing: the collected multiple source data are preprocessed, cleaning operation, conversion operation and normalization operation are performed on the preprocessed multiple source data.
[0106] The cleaning operation is used to remove repeated data, data with too many missing values and obviously incorrect data, the conversion operation is used to convert data in different formats into a unified predetermined format, the normalization operation is used to standardize data in different magnitudes, and natural language processing is performed on text form data, the natural language processing comprising word segmentation and stop word removal.
[0107] As a specific implementation of data preprocessing, the collected multi-source data is cleaned, converted and normalized. Cleaning data is to remove duplicate data, data with too many missing values and obviously incorrect data; converting data is to uniformly convert data of different formats into a format that can be recognized and processed by the system; normalization processing is to standardize data of different magnitudes so that the data is in the same range, facilitating subsequent analysis. At the same time, natural language processing operations such as word segmentation and stop word removal are performed on text data (such as user reviews) to extract key information.
[0108] Step S300 data fusion: the preprocessed multi-source data is fused to form a feature vector including hotel price related information.
[0109] As a specific implementation of data fusion, the preprocessed data is fused to construct a multi-dimensional hotel price monitoring data set. In a feature fusion manner, different sources and different types of data features are integrated to form a feature vector containing multiple aspects of information related to hotel prices, providing comprehensive data support for subsequent analysis.
[0110] Step S400 model training: based on a machine learning method, a price prediction model is constructed, and the price prediction model is trained and tested based on the feature vector to obtain a trained price prediction model.
[0111] In this embodiment, model training includes the following operations:
[0112] (1) Data set division: the feature vector is divided into a training set and a test set by stratified sampling;
[0113] (2) Feature engineering: for continuous data in the training set and the test set, the data is standardized to obtain processed features; for discrete data in the training set and the test set, the data is one-hot encoded to obtain processed features; for textual data including user reviews, the data is converted into numerical features to obtain processed features, and based on the feature importance ranking of the random forest, features that have an impact on price anomalies are selected from the processed features to obtain sample features;
[0114] (3) Model training: based on a machine learning method, a price prediction model is constructed, and the price prediction model is trained and tested based on the sample features to obtain a trained price prediction model.
[0115] The price prediction model is a network model constructed based on a random forest and a gradient boosting decision tree, and the model training of the price prediction model includes the following:
[0116] (1) Model training of the random forest model;
[0117] (2) Model training of gradient boosting decision tree model;
[0118] (3) Ensemble model construction: Adjust the weight through cross-validation to make the ensemble model optimal in the test set, and integrate the prediction results of the two models by weighted voting method, and weighted average according to the probability value output by the model.
[0119] The training steps of the random forest model are as follows:
[0120] (1) Bagging sampling: multiple subsets are extracted from the training set with replacement, and each subset trains a decision tree;
[0121] (2) Random selection of features: when each tree node is split, a part of features is randomly selected to calculate the best split point.
[0122] The training steps of the gradient boosting decision tree model are as follows:
[0123] (1) Initialization of prediction: use the mean value of all samples as the initial prediction value;
[0124] (2) Calculate the residual: the difference between the current model prediction value and the true value is used as the target of the next tree;
[0125] (3) Iterative training: each tree fits the residual of the previous round, and minimizes the loss function through gradient descent;
[0126] (4) Regularization: control overfitting through learning_rate and subsample.
[0127] Step S500 monitors and warns: real-time multi-source data is collected, preprocessed, and fused to form a feature vector, and the feature vector is input to determine whether the hotel price has abnormal fluctuations by using the trained price monitoring model, predict the output abnormality degree type and the probability value corresponding to the abnormality degree type as the prediction result, and send warning information based on the prediction result.
[0128] In this embodiment, the warning information further includes the abnormal hotel name, the abnormal time, the abnormal price and the abnormal reason analysis, and the abnormal reason analysis is generated based on the prediction result.
[0129] The monitoring and warning are in the form of short message, email and APP push to push the warning information.
[0130] The step inputs the data collected in real time and after pretreatment and fusion into the trained machine learning model, the model analyzes and predicts the input data, and judges whether the hotel price has abnormal fluctuation. When the model judges that the hotel price has abnormal fluctuation, the system sends early warning information to the relevant management department and consumers through various ways such as short message, email, APP push and the like according to the abnormal degree. At the same time, the system generates a detailed abnormal fluctuation report including the abnormal hotel name, abnormal time, abnormal price, abnormal reason analysis (based on multi-source data feature analysis) and the like, which provides a basis for the management department to supervise and for the consumers to make decisions.
[0131] The above describes the hotel price monitoring system and method based on multi-data source and machine learning provided by the application in detail, the principle and implementation mode of the application are described by applying specific examples in this paper, the above example is only used to help understand the method and core idea of the application; at the same time, for the general technical personnel in the art, according to the idea of the application, the specific implementation mode and application range will be changed, and the above description should not be understood as the limitation of the application.
Claims
1. A hotel price monitoring system based on multi-data sources and machine learning, characterized in that, The method comprises a data collection module, a data preprocessing module, a data fusion module, a model training module, and a monitoring and early warning module. The data collection module is configured to collect multi-source data from multiple tourism platforms and tourism management department databases based on a web crawler technology. The data preprocessing module is configured to preprocess the collected multi-source data, perform cleaning, conversion, and normalization operations on the multi-source data, and obtain preprocessed multi-source data. The data fusion module is configured to fuse the preprocessed multi-source data to form a feature vector comprising hotel price-related information. The model training module is configured to construct a price prediction model based on a machine learning method, train and test the price prediction model based on the feature vector, and obtain a trained price prediction model. The monitoring and early warning module is configured to collect real-time multi-source data using the data collection module, preprocess the real-time multi-source data using the data preprocessing module, obtain preprocessed multi-source data, fuse the preprocessed multi-source data using the data fusion module to form a feature vector, input the feature vector into the trained price monitoring model to determine whether the hotel price has abnormally fluctuated, predict the type of abnormality and the probability value corresponding to the type of abnormality as the prediction result, and send an early warning message based on the prediction result.
2. The multi-data source and machine learning based hotel price monitoring system as claimed in claim 1, wherein, The multi-source data includes hotel name, hotel location, room type information, price, occupancy rate, user evaluation, holiday tourist flow prediction data, and local hotel industry average price, wherein the price includes current price, historical price, and holiday price. When collecting multi-source data based on a web crawler technology from multiple tourism platforms and tourism management department databases, the web crawler sets the access frequency and request header information as needed.
3. The multi-data source and machine learning based hotel price monitoring system as claimed in claim 1, wherein, The cleaning operation removes duplicate data, data with excessive missing values, and obviously incorrect data, the conversion operation converts data in different formats into a unified predetermined format, the normalization operation standardizes data of different magnitudes, and natural language processing is performed on text data, including word segmentation and stop word removal.
4. The multi-data source and machine learning based hotel price monitoring system, as claimed in claim 1, wherein, The model training module is configured to perform the following operations: Data set division: divide the feature vector into a training set and a test set by stratified sampling; Feature engineering: for continuous data in the training set and the test set, perform standardization processing to obtain processed features, for discrete data in the training set and the test set, perform one-hot encoding to obtain processed features, for textual data including user evaluations, convert the data into numerical features to obtain processed features, and based on the feature importance ranking of a random forest, select features that have an impact on price anomalies from the processed features to obtain sample features; Model training: construct a price prediction model based on a machine learning method, train and test the price prediction model based on the sample features, and obtain a trained price prediction model; The price prediction model is a network model constructed based on a random forest and a gradient boosting decision tree, and the model training of the price prediction model includes the following: Train the random forest model. Model training is performed on the gradient boosting decision tree model; Model construction: the integrated model is adjusted by cross-validation to make the performance of the integrated model optimal on the test set, and the prediction results of the two models are integrated by weighted voting, and the probability value output by the model is weighted and averaged; The training steps of the random forest model are as follows: Bagging sampling: multiple subsets are extracted from the training set with replacement, and each subset trains a decision tree; Feature random selection: when each tree node is split, a part of features is randomly selected to calculate the best split point; The training steps of the gradient boosting decision tree model are as follows: Initialization prediction: the mean value of all samples is used as the initial prediction value; Calculate the residual: the difference between the current model prediction value and the true value is used as the target of the next tree; Iterative training: each tree fits the residual of the previous round, and the loss function is minimized by gradient descent; Regularization: control overfitting by learning_rate and subsample.
5. The multi-data source and machine learning based hotel price monitoring system, as claimed in claim 1, wherein, The early warning information also includes abnormal hotel name, abnormal time, abnormal price and abnormal reason analysis, which is generated based on the prediction result; The monitoring and early warning module is used to push the early warning information in the form of SMS, email and APP push.
6. A hotel price monitoring method based on multi-data source and machine learning, characterized in that, Comprising the following steps: Data acquisition: based on network crawler technology, collect multi-source data from multiple tourism platforms and tourism management department databases; Data preprocessing: preprocessing the collected multi-source data, cleaning, converting and normalizing the data to obtain preprocessed multi-source data; Data fusion: fuse the preprocessed multi-source data to form a feature vector including hotel price related information; Model training: based on machine learning method, construct price prediction model, and based on feature vector, model training and testing of price prediction model are performed to obtain trained price prediction model; Monitoring and early warning: collect real-time multi-source data, preprocess the real-time multi-source data to obtain preprocessed multi-source data, fuse the preprocessed multi-source data to form a feature vector, input the feature vector into the trained price monitoring model to determine whether the hotel price has abnormal fluctuation, predict the output abnormal degree type and the probability value corresponding to the abnormal degree type as the prediction result, and send the early warning information based on the prediction result.
7. The multi-data source and machine learning based hotel price monitoring method of claim 6, wherein, The multi-source data includes hotel name, hotel geographical location, room type information, price, occupancy rate, user evaluation, holiday tourist flow prediction data, and local hotel industry average price, wherein the price includes current day price, historical price and holiday price; When collecting multi-source data based on network crawler technology from multiple tourism platforms and tourism management department databases, the network crawler sets the access frequency and request header information according to the needs.
8. The multi-data source and machine learning based hotel price monitoring method of claim 6, wherein, Through cleaning operation, remove duplicate data, data with too many missing values and obviously incorrect data, through conversion operation, convert data of different formats to a unified predetermined format, through normalization operation, standardize data of different magnitudes, and perform natural language processing on text data, including word segmentation and stop word removal.
9. The multi-data source and machine learning based hotel price monitoring method of claim 6, wherein, Model training includes the following operations: Dataset division: the feature vector is divided into training set and test set by stratified sampling; Feature engineering: for continuous data in the training set and the test set, the data is standardized to obtain processed features; for discrete data in the training set and the test set, the data is one-hot encoded to obtain processed features; for textual data including user reviews, the data is converted into numerical features to obtain processed features; and based on the feature importance ranking of the random forest, features that have an impact on price anomalies are selected from the processed features to obtain sample features; Model training: a price prediction model is constructed based on a machine learning method, the price prediction model is trained and tested based on the sample features, and a trained price prediction model is obtained; The price prediction model is a network model constructed based on a random forest and a gradient boosting decision tree, and the model training of the price prediction model includes the following: Model training of the random forest model; Model training of the gradient boosting decision tree model; Ensemble model construction: the weights are adjusted through cross-validation to make the ensemble model optimal in performance on the test set, the prediction results of the two models are integrated by weighted voting, and the weighted average is calculated according to the probability value output by the model; The training steps of the random forest model are as follows: Bagging sampling: multiple subsets are extracted from the training set with replacement, and each subset trains a decision tree; Random feature selection: when each tree node is split, a part of features is randomly selected to calculate the best split point; The training steps of the gradient boosting decision tree model are as follows: Initialization prediction: the mean value of all samples is used as the initial prediction value; Residual calculation: the difference between the current model prediction value and the true value is used as the target of the next tree; Iterative training: each tree fits the residual of the previous round, and the loss function is minimized by gradient descent; Regularization: learning_rate and subsample are used to control overfitting.