User experience quality assessment method and system based on incremental learning and dynamic updating

Abstract

The invention discloses a user experience quality evaluation method and system based on incremental learning and dynamic updating. The method comprises the steps of original data acquisition, data preprocessing, data mining and construction of a wide forest-based user QoE prediction model. When the constructed QoE prediction model cannot reach the performance required in practical application, the existing wide forest is dynamically updated through incremental learning, and under the condition that user data is continuously increased, the weight is updated only for new input through the incremental learning mode.

Description

Technical Field

[0001] The present invention relates to the field of user experience quality assessment, and in particular to a method and system for user experience quality assessment based on incremental learning and dynamic updating. Background Art

[0002] As the user scale grows, users' requirements for service experience are getting higher and higher. In order to improve user stickiness and market competitiveness, enterprises or operators are in urgent need of a suitable quality evaluation system to reflect and characterize users' evaluation and experience of the current service quality. Quality of Experience (QoE) is defined as "the overall acceptability of the end user to the application or service." For example, in the IPTV scenario, the factors affecting QoE include not only network-related objective factors such as bit rate, delay, and packet loss rate, but also the subjective factors of users and the factors caused by the user's environment. In recent years, user-centric QoE has become the preferred strategy for evaluating operator business evaluation indicators. By actively obtaining user QoE, operators can help grasp user needs, improve service quality, and enhance market competitiveness.

[0003] The traditional way to obtain QoE is through user satisfaction surveys, but this places high demands on users and lacks real-time performance, which does not meet the needs of operators. Therefore, it is necessary to study the evaluation method for QoE to achieve real-time and accurate analysis and prediction of user QoE, thereby improving services and ultimately enhancing user experience.

[0004] Therefore, there is a need for methods and systems that can improve upon the deficiencies in the prior art. Summary of the invention

[0005] This Summary is provided to introduce some concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

[0006] In response to the technical problems in the prior art, the technical solution in the present disclosure is based on the theoretical background of machine learning algorithms and provides a method for evaluating the quality of user experience in a big data environment, thereby realizing real-time and accurate analysis and prediction of user QoE, thereby improving services and ultimately enhancing user experience.

[0007] Specifically, in one embodiment of the present invention, a method for evaluating user experience quality (QoE) is provided, the method comprising:

[0008] Get user data;

[0009] Extract key parameters from the acquired user data;

[0010] A user QoE prediction model based on a width forest is constructed by taking these key parameters as input data, and the width forest includes a feature layer forest and an enhancement layer forest;

[0011] Adaptively and iteratively expand the enhancement layer forest to strengthen the user QoE prediction model when the user QoE prediction model cannot achieve the required performance;

[0012] Continuously acquiring new user data and updating the enhanced user QoE prediction model based on the new user data; and

[0013] The user data to be predicted is input into the updated and enhanced user QoE prediction model to obtain the final user QoE prediction result.

[0014] In one embodiment of the present invention, acquiring user data further includes cleaning and standardizing the acquired user data.

[0015] In one embodiment of the present invention, extracting key parameters from the acquired user data further includes extracting key parameters that affect user QoE using data mining technology, wherein the key parameters are extracted by comparing the relationship between data and the target using a Pearson correlation coefficient method.

[0016] In one embodiment of the present invention, constructing the user QoE prediction model based on width forest further includes: constructing the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest.

[0017] In the above embodiment of the present invention, constructing the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest further comprises:

[0018] Initialize each forest;

[0019] Training the feature layer forest based on the input data to generate corresponding prediction probabilities and compressing the input data into important feature data;

[0020] Using the important feature data and the prediction probability as inputs of the enhancement layer forest to train the enhancement layer forest; and

[0021] The probability matrix of the class distribution estimated by the feature layer forest and the enhancement layer forest is combined as the input of the final layer and the connection weight is determined by taking the pseudo-inverse of the input.

[0022] In one embodiment of the present invention, adaptively and iteratively expanding the enhancement layer forest further comprises:

[0023] Based on the new output of the expanded enhancement layer forest and the mapping relationship between the new output and the new input of the expanded enhancement layer forest, pseudo-inverse the new input to update the connection weights of the enhancement layer forest;

[0024] Verify whether the expanded enhancement layer forest brings significant performance improvement by comparing the average out-of-bag errors of all forest groups; and

[0025] The enhancement layer forest is continued to be expanded when the out-of-bag error is significantly reduced and the enhancement layer forest is stopped to be expanded when the out-of-bag error is almost unchanged.

[0026] In the above-mentioned embodiments of the present invention, the feature layer forest is expanded in the same manner when the expanded enhancement layer forest cannot achieve the required performance.

[0027] In one embodiment of the present invention, updating the enhanced user QoE prediction model based on the new user data further includes:

[0028] The new user data is used as input of the feature layer forest and the enhancement layer forest to obtain output;

[0029] representing a new input based on the output; and

[0030] The new input is added to the input matrix of the width forest and the pseudo-inverse of the input matrix is ​​calculated to update the connection weights of the width forest.

[0031] In another embodiment of the present invention, a system for evaluating user experience quality (QoE) is provided, the system comprising:

[0032] A user data acquisition module configured to acquire user data;

[0033] A key parameter extraction module, configured to extract key parameters from the acquired user data;

[0034] A user QoE prediction model building module is configured to build a user QoE prediction model based on a width forest by taking the key parameters as input data, the width forest including a feature layer forest and an enhancement layer forest;

[0035] The user QoE prediction model enhancement and update module is configured to:

[0036] Adaptively and iteratively expand the enhancement layer forest to strengthen the user QoE prediction model when the user QoE prediction model fails to achieve the required performance; and

[0037] Continuously acquiring new user data and updating the enhanced user QoE prediction model based on the new user data; and

[0038] The user QoE prediction module is configured to input the to-be-predicted user data into the updated and enhanced user QoE prediction model to obtain a final user QoE prediction result.

[0039] In one embodiment of the present invention, the user QoE prediction model construction module is further configured to construct the width forest-based user QoE prediction model by constructing the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest, and the user QoE prediction model construction module is further configured to construct the feature layer forest and the enhancement layer forest and estimate the connection weights of the feature layer forest and the enhancement layer forest by the following operations:

[0040] Initialize each forest;

[0041] Training the feature layer forest based on the input data to generate corresponding prediction probabilities and compressing the input data into important feature data;

[0042] Using the important feature data and the prediction probability as inputs of the enhancement layer forest to train the enhancement layer forest; and

[0043] The probability matrix of the class distribution estimated by the feature layer forest and the enhancement layer forest is combined as the input of the final layer and the connection weight is determined by taking the pseudo-inverse of the input.

[0044] In one embodiment of the present invention, the user QoE prediction model enhancement and update module is further configured to adaptively and iteratively expand the enhancement layer forest by:

[0045] Based on the new output of the expanded enhancement layer forest and the mapping relationship between the new output and the new input of the expanded enhancement layer forest, pseudo-inverse the new input to update the connection weights of the enhancement layer forest;

[0046] Verify whether the expanded enhancement layer forest brings significant performance improvement by comparing the average out-of-bag errors of all forest groups; and

[0047] The enhancement layer forest is continued to be expanded when the out-of-bag error is significantly reduced and the enhancement layer forest is stopped to be expanded when the out-of-bag error is almost unchanged.

[0048] In one embodiment of the present invention, the user QoE prediction model enhancement and update module is further configured to update the enhanced user QoE prediction model based on the new user data by:

[0049] The new user data is used as input of the feature layer forest and the enhancement layer forest to obtain output;

[0050] representing a new input based on the output; and

[0051] The new input is added to the input matrix of the width forest and the pseudo-inverse of the input matrix is ​​calculated to update the connection weights of the width forest.

[0052] In yet another embodiment of the present invention, a computer readable medium storing computer executable instructions is provided, the instructions comprising:

[0053] Instructions for obtaining user data;

[0054] Instructions for extracting key parameters from the acquired user data;

[0055] Instructions for constructing a user QoE prediction model based on a width forest by taking the key parameters as input data, the width forest including a feature layer forest and an enhancement layer forest;

[0056] Instructions for adaptively and iteratively expanding the enhancement layer forest to strengthen the user QoE prediction model when the user QoE prediction model fails to achieve the required performance;

[0057] Instructions for continuously acquiring new user data and updating the enhanced user QoE prediction model based on the new user data; and

[0058] Instructions for inputting the to-be-predicted user data into the updated and enhanced user QoE prediction model to obtain a final user QoE prediction result.

[0059] After studying the description of the specific exemplary embodiments of the present invention below in conjunction with the accompanying drawings, other aspects, features and embodiments of the present invention will be apparent to those of ordinary skill in the art. Although features of the present invention may be discussed below with respect to certain embodiments and accompanying drawings, all embodiments of the present invention may include one or more of the advantageous features discussed herein. In other words, although one or more embodiments may be discussed to have certain advantageous features, one or more of such features may also be used according to the various embodiments of the present invention discussed herein. In a similar manner, although exemplary embodiments may be discussed below as device, system or method embodiments, it should be appreciated that such exemplary embodiments may be implemented in various devices, systems, and methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to understand in detail the manner in which the above-stated features of the present disclosure are used, a more particular description of the content briefly summarized above may be made with reference to various aspects, some of which are illustrated in the accompanying drawings. It should be noted, however, that the accompanying drawings illustrate only certain typical aspects of the present disclosure and are not to be considered limiting of its scope, as the description may admit to other equally effective aspects.

[0061] Figure 1 A schematic block diagram of a system for user QoE evaluation according to an embodiment of the present disclosure is shown.

[0062] Figure 2 A schematic diagram of dynamically increasing an enhancement layer forest based on adaptive incremental learning according to an embodiment of the present disclosure is shown.

[0063] Figure 3 A schematic diagram of dynamically increasing a feature layer forest based on adaptive incremental learning according to an embodiment of the present disclosure is shown.

[0064] Figure 4 A schematic diagram of adding input user data according to an embodiment of the present disclosure is shown.

[0065] Figure 5 A flowchart of a method for user QoE evaluation according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0066] The various embodiments will be described in more detail below with reference to the accompanying drawings that form part of the present invention and illustrate various specific exemplary embodiments. However, the various embodiments can be implemented in many different forms and should not be interpreted as limiting the various embodiments set forth herein; on the contrary, these embodiments are provided to make the present disclosure thorough and complete, and to fully convey the scope of these embodiments to those of ordinary skill in the art. The various embodiments may be implemented in accordance with methods, systems, or devices. Therefore, these embodiments may be implemented in hardware, in full software, or in a combination of software and hardware. Therefore, the following specific embodiments are not limiting.

[0067] The steps in each flow chart may be performed by hardware (e.g., processor, engine, memory, circuit), software (e.g., operating system, application, driver, machine / processor executable instructions), or a combination thereof. As will be appreciated by those skilled in the art, the methods involved in each embodiment may include more or fewer steps than shown.

[0068] In view of the defects in the prior art, the present invention proposes a user experience quality evaluation method based on big data incremental learning. The method includes raw data collection, data preprocessing, data mining, and construction of a user QoE prediction model based on width forest. By using data mining methods, the key parameters affecting user QoE are extracted to reduce the complexity of the model; the width forest algorithm is applied to the user experience quality QoE evaluation to improve the accuracy of predicting user QoE; and further considering that the amount of user data under the background of big data is large and complex, when the width forest algorithm cannot achieve the required performance in practical applications, the idea of ​​incremental learning is used to transform the existing width forest, through two means: 1. Dynamically increase the enhancement layer forest based on adaptive incremental learning; 2. Dynamically increase the feature layer forest based on adaptive incremental learning to adaptively tune the model, and further improve the accuracy of user QoE prediction under the complex background of big data. In addition, the situation of continuous increase of user data is also considered. The weights are updated only for new inputs through incremental learning, instead of retraining the entire model, saving computing costs and time consumption.

[0069] The various aspects of the present disclosure will be described in more detail and more comprehensively through block diagrams and method flow charts below.

[0070] Figure 1 FIG. 1 is a schematic block diagram of a system 100 for user QoE evaluation according to an embodiment of the present disclosure.

[0071] like Figure 1 As shown, the system 100 may include a user data acquisition module 102 , a key parameter extraction module 104 , a user QoE prediction model construction module 106 , a user QoE prediction model enhancement and update module 108 , and a user QoE prediction module 110 .

[0072] In one embodiment of the present invention, user data acquisition module 102 can be configured to obtain user data. As an example and not a limitation, in the scene of IPTV, means such as set-top box probes can be utilized to collect IPTV user viewing behavior and set-top box network status data. In this scene, user viewing records can be collected at intervals of every 5 minutes by means of set-top box probes, including behavioral data such as user switching, pausing, and network status information data such as delays and jitters. In one embodiment of the present invention, obtaining user data can further include cleaning and standardizing the user data obtained. In this embodiment, the collected data can be cleaned, including de-duplication, deleting meaningless data such as empty attributes, and utilizing methods such as Z-Score zero mean normalization to standardize the cleaned data, as shown in the following formula:

[0073]

[0074] As can be appreciated by those skilled in the art, in other embodiments, any other suitable normalization method may be used to normalize the cleaned data, without being limited to the above method.

[0075] In one embodiment of the present invention, the key parameter extraction module 104 may be configured to extract key parameters from the acquired user data. In this embodiment, extracting key parameters from the acquired user data may further include extracting key parameters that affect user QoE using data mining technology. Specifically, as an example and not a limitation, the Pearson correlation coefficient method may be used to extract key factors that affect user QoE from the acquired user data by comparing the relationship between data and the target. The stronger the correlation, the closer the correlation coefficient r is to 1, so as to mine the user data features. The formula is as follows:

[0076]

[0077] In one embodiment of the present invention, the user QoE prediction model construction module 106 may be configured to construct a user QoE prediction model based on a width forest by using these key parameters as input data. In one embodiment of the present invention, the width forest includes a feature layer forest and an enhancement layer forest. In this embodiment, the user QoE prediction model construction module 106 may be further configured to construct a user QoE prediction model based on a width forest by the following operations: constructing the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest. That is, the extracted key parameters are used as input data, and the training data set and the test data set are divided for constructing the user QoE prediction model. First, a width forest is constructed based on the training data, and the process includes the construction of the feature layer forest, the construction of the enhancement layer forest, and the estimation of the connection weights.

[0078] Specifically, in one embodiment of the present invention, the user QoE prediction model building module 106 may be further configured to build the feature layer forest and the enhancement layer forest and estimate the connection weights of the feature layer forest and the enhancement layer forest by the following operations:

[0079] First, each forest is initialized. As an example but not a limitation, each forest can be initialized to 50 trees and the depth of the tree is not limited. The feature layer forest can be composed of n groups of forests, each group of forests contains two different random forests according to the different training sample extraction methods, namely, random forest (randomly extract samples) and extreme random forest (use all training samples).

[0080] Subsequently, the feature layer forest can be trained based on the input data to generate corresponding prediction probabilities and compress the input data into important feature data. Specifically, the IPTV user data input training set can be denoted as X∈RN×M , Y∈R N×K , where N represents the sample size, M represents the number of input features, and K represents the number of categories. The feature layer forest can generate n sets of prediction probabilities, expressed as Z n = {Z i} i＝1,...,n , where Z i ∈R N×2K , Z n ∈R N×2nK At the same time, the feature layer random forest can be used to select rM features with the highest importance scores (r is the feature selection ratio, which can be set to 0.8, for example), that is, the input data X∈R N×M is compressed to X'∈R N×γM .

[0081] Then, the important feature data and the predicted probability can be used as the input of the enhancement layer forest to train the enhancement layer forest. Specifically, the enhancement layer forest can be composed of m groups of random forests, which are the same as the feature layer forest, and each group contains a random forest and an extreme random forest. N×γM And the prediction output Z of the feature layer forest n , as the input of the enhancement layer forest, expressed as [Z n |X']∈R N×(2nK+γM) , after enhanced layer forest training, the output H is obtained m ={H i} i＝1,...,m , where H i ∈R N×2K , H m ∈R N×2mK .

[0082] Finally, the probability matrix of the feature layer forest and the enhancement layer forest for class distribution estimation can be combined as the input of the final layer and the connection weight can be determined by taking the pseudo-inverse of the input. Specifically, the probability matrix of the feature layer forest and the enhancement layer forest for class distribution estimation is combined as the input of the final layer, expressed as [Z n |H m ]∈R N×(2nK+2mK) , the algorithm output is expressed as: in, Represents the connection weight of the width structure; by input data [Z n |H m ] Find the pseudo-inverse to calculate the connection weight Right now: [Z n |H m ] + =([Z n |H m ]·[Zn |H m ] T +λI) -1 ·[Z n |H m ] T , where λ is the regularization parameter (which can be set to 10 by default, for example -3 ).

[0083] In the context of big data, due to the large amount and complexity of user data, the above-mentioned width forest may not be able to achieve the required performance, so it is necessary to consider adaptive tuning of the algorithm. At the same time, considering that user data will continue to be input over time, traditional algorithm models usually put new data and old trained data together, and retraining the entire algorithm model requires a lot of time and space. The idea of ​​incremental learning is to update the trained model based on the changes caused by new data or newly added structures of the model, rather than retraining. It is suitable for complex application scenarios of big data and saves computing resources while ensuring the accuracy of the model. This patent uses the idea of ​​incremental learning to transform the existing width forest based on a dynamically and gradually updated algorithm to improve the accuracy of user QoE prediction under big data.

[0084] Therefore, in one embodiment of the present invention, the user QoE prediction model strengthening and updating module 108 may be configured to adaptively iteratively expand the enhancement layer forest to strengthen the user QoE prediction model when the user QoE prediction model cannot achieve the required performance. In this embodiment of the present invention, the user QoE prediction model strengthening and updating module 108 may be further configured to adaptively iteratively expand the enhancement layer forest by the following operations:

[0085] First, based on the new output of the expanded enhancement layer forest and the mapping relationship between the new output and the new input of the expanded enhancement layer forest, the new input can be pseudo-inverted to update the connection weights of the enhancement layer forest. Specifically, Figure 2 FIG. 2 shows a schematic diagram of dynamically increasing the enhancement layer forest based on adaptive incremental learning according to an embodiment of the present disclosure. Figure 2 As shown, the input of the constructed width forest is recorded as After the enhancement layer forest is expanded, the expanded forest obtains a new output, which is recorded as Among them, f(·) represents the mapping relationship between the input and output of the enhancement layer forest, and f(Z n |X′) represents the output of the newly added enhancement layer forest. According to the pseudo-inverse principle of the partition matrix in the dynamic step-by-step update algorithm, the pseudo-inverse of the new input of the width forest is obtained:

[0086]

[0087] In order to facilitate the above expression, D and B in the formula T for:

[0088]

[0089]

[0090]

[0091] Will Denote the connection weight of the width forest before updating as The updated weight is expressed as

[0092]

[0093] Then, the average out-of-bag errors of all forest groups can be compared to verify whether the expanded enhancement layer forest brings significant performance improvement. If the out-of-bag error is significantly reduced, the enhancement layer forest will continue to be expanded, and if the out-of-bag error is almost unchanged (technical personnel in the field can set a change threshold to define what is almost unchanged), the enhancement layer forest will stop expanding.

[0094] In another embodiment of the present invention, the user QoE prediction model enhancement and updating module 108 may be further configured to expand the feature layer forest in the same or similar manner if the expanded enhancement layer forest fails to achieve the required performance.

[0095] Specifically, Figure 3 FIG. 1 shows a schematic diagram of dynamically increasing feature layer forests based on adaptive incremental learning according to an embodiment of the present disclosure. Figure 3 As shown in Figure 2, adding feature layer forests can allow more forests to participate in feature selection decisions, and utilizing the representation capabilities of more forests is conducive to extracting more effective features. The predicted output of the added feature layer forest is denoted as Z n+1 , then the output of the corresponding added enhancement layer forest is f(Z n+1 |X′). Therefore, the input of the updated width forest is Seek the pseudo-inverse

[0096]

[0097] Among them, D and B in the formula T for:

[0098]

[0099]

[0100] The updated connection weight is recorded as Then:

[0101]

[0102] Therefore, in one embodiment of the present invention, the user QoE prediction model strengthening and updating module 108 may also be configured to continuously obtain new user data and update the strengthened user QoE prediction model based on the new user data. In this embodiment, the user QoE prediction model strengthening and updating module 108 may be further configured to update the strengthened user QoE prediction model based on the new user data by the following operations:

[0103] First, the new user data can be used as the input of the feature layer forest and the enhancement layer forest to obtain the output. Specifically, Figure 4 FIG. 2 shows a schematic diagram of adding input user data according to an embodiment of the present disclosure. Figure 4 As shown, assuming that the new input training data is {X a ,Y a}, the data dimension is N a , the outputs of the existing model, i.e., the n-group feature layer forest and m-group enhancement layer forest constructed above, are expressed as and Then, a new input can be represented based on the output. Specifically, the new input can be represented as A a =[Z a n |H a m ]. Finally, the new input can be added to the input matrix of the width forest and the pseudo-inverse of the input matrix can be calculated to update the connection weights of the width forest. Specifically, after adding the new input, the input matrix of the width forest becomes:

[0104]

[0105] According to the pseudo-inverse principle of matrix partition, we can find the pseudo-inverse:

[0106]

[0107] Where D and B in the formula T Respectively expressed as:

[0108]

[0109]

[0110]

[0111] Accordingly, its updated weight is expressed as:

[0112]

[0113] In one embodiment of the present invention, the user QoE prediction module 110 may be configured to input the to-be-predicted user data into the updated and enhanced user QoE prediction model to obtain a final user QoE prediction result.

[0114] Figure 5 A flow chart of a method 500 for user QoE evaluation according to an embodiment of the present disclosure is shown.

[0115] like Figure 5 As shown, method 500 starts at step 502, obtaining user data. In one embodiment of the present invention, obtaining user data may further include cleaning and standardizing the obtained user data.

[0116] Then, the method 500 proceeds to step 504, extracting key parameters from the acquired user data. In one embodiment of the present invention, extracting key parameters from the acquired user data may further include extracting key parameters that affect user QoE using data mining technology, and the key parameters may be extracted by comparing the relationship between data and the target using the Pearson correlation coefficient method.

[0117] Next, the method 500 proceeds to step 506, and a user QoE prediction model based on a width forest is constructed by taking these key parameters as input data, and the width forest includes a feature layer forest and an enhancement layer forest. In one embodiment of the present invention, constructing a user QoE prediction model based on a width forest further includes: constructing the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest. In the above embodiment of the present invention, constructing the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest further includes: initializing each forest; training the feature layer forest based on the input data to generate corresponding prediction probabilities and compressing the input data into important feature data; training the enhancement layer forest using the important feature data and the prediction probability as inputs of the enhancement layer forest; and combining the probability matrix of the feature layer forest and the enhancement layer forest for class distribution estimation as the input of the final layer and determining the connection weight by taking the pseudo-inverse of the input.

[0118] Then, the method 500 proceeds to step 508, where the enhancement layer forest is adaptively and iteratively expanded to strengthen the user QoE prediction model when the user QoE prediction model cannot achieve the required performance. In one embodiment of the present invention, adaptively and iteratively expanding the enhancement layer forest may further include: based on the new output of the expanded enhancement layer forest and the mapping relationship between the new output and the new input of the expanded enhancement layer forest, pseudo-inverse the new input to update the connection weights of the enhancement layer forest; verifying whether the expanded enhancement layer forest brings significant performance improvement by comparing the average out-of-bag errors of all forest groups; and continuing to expand the enhancement layer forest when the out-of-bag error is significantly reduced and stopping expanding the enhancement layer forest when the out-of-bag error is almost unchanged. In the above embodiment of the present invention, the feature layer forest is expanded in the same or similar manner when the expanded enhancement layer forest cannot achieve the required performance.

[0119] Next, the method 500 proceeds to step 510, continuously acquiring new user data and updating the enhanced user QoE prediction model based on the new user data. In one embodiment of the present invention, updating the enhanced user QoE prediction model based on the new user data may further include: using the new user data as input of the feature layer forest and the enhancement layer forest to obtain an output; representing a new input based on the output; and adding the new input to the input matrix of the width forest and obtaining a pseudo-inverse of the input matrix to update the connection weights of the width forest.

[0120] Finally, the method 500 proceeds to step 512 to input the to-be-predicted user data into the updated and enhanced user QoE prediction model to obtain a final user QoE prediction result.

[0121] Method 500 ends after step 512 .

[0122] In summary, the present invention proposes an IPTV user experience quality assessment method based on big data incremental learning, and applies the width forest algorithm and the idea of ​​incremental learning to the IPTV user experience quality assessment. Considering that the amount of IPTV user data under the background of big data is large and complex, when the width forest algorithm cannot achieve the required performance in practical applications, the idea of ​​incremental learning is used to transform the existing width forest, that is, through two means: 1. Dynamically increase the enhancement layer forest based on adaptive incremental learning, 2. Dynamically increase the feature layer forest based on adaptive incremental learning, so as to adaptively tune the model and further improve the accuracy of user QoE prediction under the complex background of big data. At the same time, the situation of continuous increase of user IPTV data is taken into account. The weights are updated only for new inputs through incremental learning, instead of retraining the entire model, which saves computing costs and time consumption.

[0123] Embodiments of the present invention have been described above with reference to block diagrams and / or the operating instructions of methods, systems, and computer program products according to embodiments of the present invention. Each function / action noted in the frame may occur in an order different from that shown in any flow chart. For example, depending on the function / action involved, two frames shown in succession may actually be performed substantially simultaneously, or these frames may sometimes be performed in an opposite order.

[0124] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for evaluating user experience quality (QoE), the method comprising: Get user data; Extract key parameters from the acquired user data; A user QoE prediction model based on a width forest is constructed by taking these key parameters as input data, wherein the width forest includes a feature layer forest and an enhancement layer forest; Adaptively and iteratively expanding the enhancement layer forest to strengthen the user QoE prediction model when the user QoE prediction model fails to achieve the required performance; Continuously acquiring new user data and updating the enhanced user QoE prediction model based on the new user data; as well as The user data to be predicted is input into the updated and enhanced user QoE prediction model to obtain the final user QoE prediction result.

2. The method as claimed in claim 1, wherein extracting key parameters from the acquired user data further comprises using data mining technology to extract key parameters that affect user QoE, and wherein the key parameters are extracted by comparing the relationship between data and targets using a Pearson correlation coefficient method.

3. The method of claim 1, wherein constructing a user QoE prediction model based on a width forest further comprises: The feature layer forest and the enhancement layer forest are constructed, and connection weights of the feature layer forest and the enhancement layer forest are estimated.

4. The method of claim 3, wherein constructing the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest further comprises: Initialize each forest; Training the feature layer forest based on the input data to generate corresponding prediction probabilities and compressing the input data into important feature data; Training the enhancement layer forest by using the important feature data and the predicted probability as inputs of the enhancement layer forest; as well as The probability matrices of class distribution estimates from the feature layer forest and the enhancement layer forest are combined as the input of the final layer and the connection weights are determined by taking the pseudo-inverse of the input.

5. The method of claim 1, wherein adaptively and iteratively expanding the enhancement layer forest further comprises: Based on the new output of the expanded enhancement layer forest and the mapping relationship between the new output and the new input of the expanded enhancement layer forest, pseudo-inverse the new input to update the connection weights of the enhancement layer forest; Verify whether the extended enhancement layer forest brings significant performance improvement by comparing the average out-of-bag errors of all forest groups; as well as The enhancement layer forest is continued to be expanded when the out-of-bag error is significantly reduced and the enhancement layer forest is stopped to be expanded when the out-of-bag error is almost unchanged.

6. The method of claim 5, wherein the feature layer forest is expanded in the same manner if the expanded enhancement layer forest fails to achieve the desired performance.

7. The method of claim 1, wherein updating the enhanced user QoE prediction model based on the new user data further comprises: Taking the new user data as input of the feature layer forest and the enhancement layer forest to obtain output; representing a new input based on the output; as well as The new input is added to the input matrix of the width forest and the pseudo-inverse of the input matrix is ​​calculated to update the connection weights of the width forest.

8. A system for evaluating user experience quality (QoE), the system comprising: A user data acquisition module configured to acquire user data; A key parameter extraction module, configured to extract key parameters from the acquired user data; A user QoE prediction model building module is configured to build a user QoE prediction model based on a width forest by taking the key parameters as input data, wherein the width forest includes a feature layer forest and an enhancement layer forest; The user QoE prediction model enhancement and update module is configured to: Adaptively and iteratively expanding the enhancement layer forest to strengthen the user QoE prediction model when the user QoE prediction model fails to achieve the required performance; and Continuously acquiring new user data and updating the enhanced user QoE prediction model based on the new user data; as well as The user QoE prediction module is configured to input the to-be-predicted user data into the updated and enhanced user QoE prediction model to obtain a final user QoE prediction result.

9. The system of claim 8, wherein the user QoE prediction model building module is further configured to build a width forest-based user QoE prediction model by building the feature layer forest and the enhancement layer forest and estimating the connection weights of the feature layer forest and the enhancement layer forest, and wherein the user QoE prediction model building module is further configured to build the feature layer forest and the enhancement layer forest and estimate the connection weights of the feature layer forest and the enhancement layer forest by: Initialize each forest; Training the feature layer forest based on the input data to generate corresponding prediction probabilities and compressing the input data into important feature data; Training the enhancement layer forest by using the important feature data and the predicted probability as inputs of the enhancement layer forest; as well as The probability matrices of class distribution estimates from the feature layer forest and the enhancement layer forest are combined as the input of the final layer and the connection weights are determined by taking the pseudo-inverse of the input.

10. The system of claim 8, wherein the user QoE prediction model enhancement and updating module is further configured to adaptively and iteratively expand the enhancement layer forest by: Based on the new output of the expanded enhancement layer forest and the mapping relationship between the new output and the new input of the expanded enhancement layer forest, pseudo-inverse the new input to update the connection weights of the enhancement layer forest; Verify whether the extended enhancement layer forest brings significant performance improvement by comparing the average out-of-bag errors of all forest groups; as well as Continue to expand the enhancement layer forest when the out-of-bag error is significantly reduced and stop expanding the enhancement layer forest when the out-of-bag error is almost unchanged, and The user QoE prediction model enhancement and update module is further configured to update the enhanced user QoE prediction model based on the new user data by: Taking the new user data as input of the feature layer forest and the enhancement layer forest to obtain output; representing a new input based on the output; as well as The new input is added to the input matrix of the width forest and the pseudo-inverse of the input matrix is ​​calculated to update the connection weights of the width forest.

Images