WO2012079254A1

WO2012079254A1 - Program recommending device and program recommending method

Info

Publication number: WO2012079254A1
Application number: PCT/CN2010/079958
Authority: WO
Inventors: 徐金安; 祝真宇; 满志远; 赵云龙; 尹力; 刘军
Original assignee: 北京交通大学
Priority date: 2010-12-17
Filing date: 2010-12-17
Publication date: 2012-06-21
Also published as: CN103299651A; CN103299651B

Abstract

A program recommending device and program recommending method are provided in the present invention, which belongs to the field of artificial intelligence research. The device includes: an input unit (100), a program pre-selecting unit (110), a feature extracting unit (120), a machine learning unit (130), a program forecast unit (140) and an output unit (150). The method includes: receiving language information inputted by a user; according to the language information, extracting associated electronic program information from a electronic program list database in which the electronic program information has been stored; selecting features from the extracted electronic program information to obtain feature elements, obtaining the associated information of the feature elements from a knowledge database in which language knowledge has been stored, and constructing a feature set; constructing a statistic model by using the feature set and a machine learning method; matching programs in the electronic program list database by using the statistic model; and exporting a matching result to the user. The present invention resolves a problem of cold startup of a program recommending system, ensures privacy information of a user from leakage, and improves precision, performance and practicability of program recommending.

Description

Program recommendation device and program recommendation method

The invention relates to the field of artificial intelligence research, in particular to a program recommendation device and a program recommendation method. BACKGROUND OF THE INVENTION With the rapid development of network technology, digital television and communication technologies, cable digital television, digital digital television, satellite digital television and wireless digital television have been developed to a large-scale practical stage. Digital technology has brought about a large increase in TV channels, although EPG (Electronic Program Guide) has brought convenience to people because of the book

Due to the increasing variety of program resources, it is difficult for people to quickly find the programs they really like.

In order to solve the problem of information overload, a variety of program recommendation systems have emerged. The existing program recommendation methods mainly include the following: rule-based recommendation methods, content filtering-based recommendation methods, collaborative filtering-based recommendation methods, and hybrid-based The recommended method of the strategy.

The rule-based recommendation method mainly uses various rules to implement the recommendation function of the program. The acquisition of rules can be divided into manually written rules or mining rules based on association rules. The advantage of this method is that the rules are simple and straightforward to make.

The recommendation method based on content filtering implements the recommendation function by comparing the description information of the program and the user. This method can be implemented by machine learning methods such as vector space model, Bayesian method, decision tree, and support vector machine (SVM). The advantage of this method is that the method is simple and can make appropriate predictions for the potential needs of the user.

The recommendation method based on collaborative filtering recommends programs according to the similarity of users. In calculating the similarity between users, this method can use various clustering and classification algorithms, such as K nearest neighbor (KNN), K-MEANs, fuzzy clustering, naive Bayes, SVM, etc. Wait to achieve. The advantage of this method is the ability to discover a portion of the new program of interest for the user.

The recommendation method based on the hybrid strategy refers to a combination of a content filtering based method and a collaborative filtering based recommendation method. The method can combine the advantages of the above two methods, and complement each other to form complementary advantages, which can improve the precision and effect of the recommendation to a certain extent.

In carrying out the process of the present invention, the inventors have found that the above prior art has at least the following disadvantages:

In the rule-based recommendation method, the subjectivity of the rules is strong, and the quality is difficult to guarantee; the increase of rules leads to the rules. Conflicts, difficulties in system management and upgrades. The recommendation method based on content filtering has low recommendation effect and inefficiency for new programs, and there is a "cold start" problem. The recommendation method based on collaborative filtering has low adaptive ability, low scalability, and can not solve the problem of "cold start" and the privacy of others. The scalability of the recommendation method based on the hybrid strategy needs to be further improved, the system has poor anti-malware scoring ability and there is still a problem of "cold start". In summary, the existing program recommendation technology can not solve the "cold start" problem of the program recommendation system well under the premise of ensuring non-infringement of the user's personal privacy, which greatly affects the recommendation accuracy and performance of the program recommendation system. Summary of the invention

In order to solve the problems in the prior art, embodiments of the present invention provide a program recommendation apparatus and a program recommendation method. The technical solution is as follows:

A program recommendation device, the device comprising:

An input unit, configured to receive language information input by a user;

a program pre-selection unit, configured to extract relevant electronic program information from an electronic program table database that has stored electronic program information according to language information received by the input unit;

a feature extraction unit, configured to perform feature selection on the electronic program information extracted by the program pre-selection unit, obtain a feature element, and acquire association information of the feature element from a knowledge base in which the language knowledge is stored, and construct a feature set; a unit, configured to construct a statistical model by using a feature set obtained by the feature extraction unit and a machine learning method;

a program prediction unit, configured to match, by using a statistical model constructed by the machine learning unit, a program in the electronic program table database;

And an output unit, configured to output a result of the matching of the program prediction unit to the user.

The program pre-selection unit includes:

And a first pre-selected sub-unit, configured to: when the language information received by the input unit is a keyword set, perform logical calculation on the keyword set to extract related electronic program information from the electronic program table database.

The program pre-selection unit includes:

a second pre-selected sub-unit, configured to perform word segmentation processing when the language information received by the input unit is a phrase or a sentence, calculate a segmentation result to obtain a spatial model of the user preference, and then calculate the spatial model and the electronic The similarity of the electronic program information in the program table database, and the related electronic program information is extracted based on the similarity.

The feature extraction unit further includes:

a feedback subunit, configured to use the feature element as a search keyword for an electronic program in the electronic program table database The information is searched and evaluated, and the result of the processing is fed back to the program preselection unit;

Correspondingly, the program preselecting unit is further configured to receive a result of the feedback subunit feedback, extract relevant electronic program information from the electronic program table database according to the feedback result, and output the information to the feature extraction unit. .

The knowledge base includes synonymous, synonymous, antisense, conceptual similarity of words, and any one or more of words, word classes, and semantic attributes.

A program recommendation method, the method comprising:

Receiving language information input by the user;

Extracting related electronic program information from an electronic program table database in which electronic program information has been stored, based on the language information;

Performing feature selection on the extracted electronic program information, obtaining feature elements, and acquiring association information of the feature elements from a knowledge base in which the language knowledge is stored, and constructing a feature set;

Constructing a statistical model using the feature set and the machine learning method;

Using the statistical model, matching programs in the electronic program table database;

The result of the matching is output to the user.

Extracting related electronic program information from the electronic program table database in which the electronic program information has been stored, according to the language information, including:

When the received language information is a keyword set, logical programming of the keyword set is performed to extract relevant electronic program information from the electronic program table database.

When the received language information is a phrase or a sentence, the word segmentation process is first performed, the word segmentation result is calculated to obtain a spatial model of the user's preference, and then the spatial model is similar to the electronic program information in the electronic program table database. Degree, extracting related electronic program information according to the similarity.

After performing feature selection on the extracted electronic program information to obtain a feature element, the method further includes:

Searching and evaluating the electronic program information in the electronic program table database by using the feature element as a search keyword, and extracting relevant electronic program information from the electronic program table database according to the result of the processing, and then Feature selection of electronic program information extracted according to the language information and electronic program information extracted according to the feature element, to obtain a new feature element;

Correspondingly, the association information of the feature element is obtained from the knowledge base of the stored language knowledge, and the feature set is constructed, including: The association information of the new feature element is obtained from the knowledge base in which the language knowledge is stored, and the feature set is constructed. The knowledge base includes synonymous, synonymous, antisense, conceptual similarity of words, and any one or more of words, word classes, and semantic attributes.

The technical solution provided by the embodiment of the present invention has the following beneficial effects: extracting relevant electronic program information from the electronic program table database according to the language information input by the user, and performing feature selection to obtain feature elements, and calling information stored in the knowledge base to feature The element is expanded to obtain a feature set of the user's interest space, and the feature set and the machine learning method are used to construct a statistical model, thereby matching the electronic program table database to output the matching result to the user, realizing the program recommendation, and solving the prior art " The cold start "problem, and improved the accuracy, performance and usability of the program recommendation. Since the above device is located at the user end, the method is also performed on the user side, and does not involve collecting personal information of the user on the network server side or the user end. Therefore, the privacy information of the user can be fully protected from leakage and the confidentiality is improved. In addition, the electronic program table database can be searched and evaluated by using the feature element as a search key, and then the program pre-selection is performed according to the processing result, thereby further expanding the user's interest and interest area and improving the accuracy of the program recommendation. DRAWINGS

1 is a structural diagram of a program recommendation apparatus according to Embodiment 1 of the present invention;

2 is a structural diagram of a program recommendation apparatus according to Embodiment 2 of the present invention;

3 is a flowchart of a program recommendation method according to Embodiment 3 of the present invention;

4 is a flow chart of a program recommendation method according to Embodiment 4 of the present invention. detailed description

In order to make the objects, the technical solutions and the advantages of the present invention more apparent, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

Example 1

Referring to FIG. 1, this embodiment provides a program recommendation apparatus, including:

The input unit 100 is configured to receive language information input by the user;

The program pre-selection unit 110 is configured to extract related electronic program information from the electronic program table database in which the electronic program information has been stored according to the language information received by the input unit 100;

The feature extraction unit 120 is configured to perform feature selection on the electronic program information extracted by the program pre-selection unit 110, obtain a feature element, and acquire association information of the feature element from the knowledge base of the stored language knowledge to construct a feature set; the machine learning unit 130 , the feature set and the machine learning method obtained by the feature extraction unit 120 are constructed Statistical model

The program prediction unit 140 is configured to match the programs in the electronic program table database by using the statistical model constructed by the machine learning unit 130;

The output unit 150 is configured to output a result matched by the program prediction unit 140 to the user.

The electronic program guide (EPG) involved in the embodiment of the present invention is not limited to the EPG of a television program, and is acceptable to any other recommendation system composed of an electronic program guide.

In this embodiment, the input unit 100 receives the language input by the user, and can be implemented in various manners, including but not limited to: a remote controller, a keyboard, a pointing device (such as a mouse), a handwritten character recognition, an optical character reader, and the like. Universal input modules, or voice input through a speech recognition system, and by reading text files or reading databases are acceptable. The input unit 100 can use any method as long as it performs processing to finally obtain input of language information. The user's input can be a keyword or a phrase or sentence describing the user's preferences.

In this embodiment, the program pre-selection unit 110 may include:

a first pre-selected sub-unit, configured to: when the language information received by the input unit 100 is a keyword set, perform logical calculation on the keyword set to extract relevant electronic program information from the electronic program table database; and/or,

a second pre-selected sub-unit, configured to perform word segmentation processing when the language information received by the input unit 100 is a phrase or a sentence, calculate a segmentation result to obtain a spatial model of the user's preference, and then calculate a spatial model and an electronic program table database. The similarity of the electronic program information, and the related electronic program information is extracted based on the similarity.

Specifically, the first pre-selected sub-unit may directly extract the program from the EPG database by using the keyword set, and logical operations such as logical AND, logical OR, logical non-logical, logical and non-operation may be used between the keywords in the keyword set. Implemented.

Specifically, the second pre-selected sub-unit can be processed by using a word-dividing tool, and then the spatial model of the user's preference can be obtained by using a method for calculating the word frequency for the word segmentation result, and then the similarity between the spatial model and the electronic program information in the EPG database is calculated, and then sorted. Recommended results.

In addition, the program pre-selection unit 110 may also provide the extracted electronic program information to the user, perform initial screening by the user, and then output the result of the user screening confirmation to the feature extraction unit 120.

The EPG database according to the embodiment of the present invention may be composed of an electronic program table according to a certain structure or semi-structured. Digital TVs, such as IPTV and cable TV, can generally provide programs that are available for two weeks from the day of viewing. The data in the EPG database can be extracted from the digital wireless television receiver or from the Internet. EPG—Generally includes program number, program name, program introduction, channel, start and end time, etc., which can be stored in the EPG database according to a certain data format. In the embodiment of the present invention, the program information accessed in the EPG may be past and present. The embodiment information of the present invention is not specifically limited. The EPG database of the present invention allows accumulating and storing EPG data of a past time, such as an electronic program of the past one year or six months or three months from the date of viewing by the user, the purpose of which is to provide the user with sufficient data space of interest selection. .

In this embodiment, the feature extraction unit 120 performs a feature selection method, including but not limited to: a feature extraction method based on a document frequency, an information gain method, an X ² statistical method, a mutual information method, and the like. Feature selection can be calculated based on feature weights, such as Boolean weights, absolute TF (Term Frequency), IDF (Inverse Document Frequency), TF-IDF (Terminal Frequency and Inverse Document Frequency), TFC (Term Frequency Count), ITC, Entropy Weight, TF-IWF, etc., which are not specifically described in this embodiment of the present invention. limited. The association information of the feature elements acquired by the feature extraction unit 120 from the knowledge base includes: attribute information such as semantics and concepts of the words, and the information may serve as a feature set of the user's interest and favorite space, thereby providing modeling for the machine learning unit 130. Data conditions and basis for judgment.

The knowledge base involved in the embodiments of the present invention includes synonymous, synonymous, antisense, similarity of concepts, and any one or more of words, word classes and semantic attributes. The knowledge base can not only include the above-mentioned attribute features such as semantics and concepts, but also can include organizational information related to the attribute characteristics. The organized information refers to the appropriate organization and management of the feature elements according to the structure of the knowledge in the knowledge base. Information, such as establishing the relationship of the concept and the semantic relationship of the envelope. Organizational management can be carried out according to the concept semantic network. At the same time, different weights can be assigned to each element according to the level of the conceptual semantic network to improve the performance of the system. The knowledge base can be built manually, or it can use existing dictionaries or semantic dictionaries. For example, WordNet in English, HowNet in Chinese, and EDR electronic dictionary in Japanese. At the same time, various synonyms, synonym electronic dictionaries, and the like can be utilized.

In this embodiment, the machine learning method used by the machine learning unit 130 is various, such as a supervised machine learning method or an unsupervised machine learning method, and a semi-supervised machine learning method, etc.; specifically, a support vector machine (SVM) is adopted. The decision tree (decision tr _ee ), the Bayesian, the maximum entropy, and the conditional random field are implemented by any one of the algorithms, and may also be implemented by using multiple construction hybrid algorithms, which are not implemented by the embodiment of the present invention. Specifically limited.

In this embodiment, the program prediction unit 140 may further perform sorting processing on the matching result, and then output the sorted result to the output unit 150. Accordingly, the output unit 150 outputs the result to the user.

In this embodiment, the output unit 150 outputs the result of the program recommendation to the user in various forms, which may be a file output, or a display output, etc., wherein the output may be output and presented to the user in a specific format, and the final expression manner. It may be in any form, such as a highlight recommendation, a voice reminder, etc., which is not specifically limited in this embodiment of the present invention. After obtaining the recommended program, the user can request to play the program he needs, thereby receiving the corresponding data stream for viewing.

In this embodiment, the feature extraction unit 120 may also use a clustering or classification algorithm before or after feature selection. The calculation may be performed by the machine learning unit 130 by using a clustering or classification algorithm before or after the statistical model is constructed, so as to further improve the accuracy of the program recommendation, which is not specifically limited in the embodiment of the present invention.

The device provided by the embodiment extracts relevant electronic program information from the electronic program table database according to the language information input by the user, and performs feature selection to obtain feature elements, and invokes information stored in the knowledge base to expand the feature element to obtain a user. A feature set of the hobby space, using the feature set and the machine learning method to construct a statistical model, thereby matching the electronic program table database output matching result to the user, realizing the program recommendation, and solving the "cold start" problem of the prior art, It also improves the accuracy, performance and usability of program recommendations. Since the above device is located at the user end, it does not involve collecting personal information of the user on the network server side or the user end. Therefore, the privacy information of the user can be fully protected from leakage and the confidentiality is improved. In addition, the electronic program table database can be searched and evaluated by using the feature element as a search key, and then the program pre-selection can be performed again according to the processing result, thereby further expanding the user's interest and interest area and improving the accuracy of the program recommendation. Example 2

On the basis of Embodiment 1, the present embodiment provides a program recommendation apparatus, including: an input unit 100, a program preselection unit 110, a feature extraction unit 120, a machine learning unit 130, a program prediction unit 140, and an output unit 150. The function of each unit is the same as that described in Embodiment 1, and the improvement on the basis of the improvement is that the feature extraction unit 120 may further include:

The feedback sub-unit 120a is configured to perform retrieval and evaluation processing on the electronic program information in the electronic program table database by using the above-mentioned feature element as a retrieval keyword, and feed back the result of the processing to the program pre-selection unit 110;

Correspondingly, the program pre-selection unit 110 is further configured to receive the feedback of the feedback sub-unit, extract relevant electronic program information from the electronic program table database according to the result of the feedback, and output the information to the feature extraction unit 120, so that the feature extraction unit 120 can Selecting feature information according to the electronic program information extracted by the language information and the electronic program information extracted according to the feature element, obtaining a new feature element, and acquiring association information of the new feature element from a knowledge base of the stored language knowledge, The feature set is constructed, so that the feature set can be expanded, and the user can select his favorite program more accurately, thereby improving the prediction accuracy of the system.

Further, the feature extraction unit 120 may further determine whether it is necessary to pre-select, and if yes, perform the above-mentioned feedback operation; otherwise, continue the execution in the manner of Embodiment 1. Among them, multiple ways can be used to determine whether it is necessary to pre-select, for example, a simple question window can be preset, whether the user needs to re-select the television program, or simultaneously output the feature element as a dynamic similar to the semantic network graphic. For the user, for the user to observe and analyze, etc., the embodiment of the present invention does not specifically limit this. The device provided by the embodiment extracts relevant electronic program information from the electronic program table database according to the language information input by the user, and performs feature selection to obtain feature elements, and invokes information stored in the knowledge base to expand the feature element to obtain a user. A feature set of the hobby space, using the feature set and the machine learning method to construct a statistical model, thereby matching the electronic program table database output matching result to the user, realizing the program recommendation, and solving the "cold start" problem of the prior art, It also improves the accuracy, performance and usability of program recommendations. Since the above device is located at the user end, the personal information of the user is not involved in the network server end or the user end. Therefore, the privacy information of the user can be fully protected from leakage and the confidentiality is improved. In addition, the electronic program table database can be searched and evaluated by using the feature element as a search keyword, and then the program pre-selection is performed again according to the processing result, thereby further expanding the user's interest and interest area and improving the accuracy of the program recommendation. Example 3

Referring to FIG. 3, this embodiment provides a program recommendation method, including:

S01: receiving language information input by the user;

S02: extract relevant electronic program information from an electronic program table database that has stored electronic program information according to the language information;

S03: performing feature selection on the extracted electronic program information to obtain a feature element;

S04: acquiring association information of the feature element from the knowledge base of the stored language knowledge, and constructing the feature set;

S05: constructing a statistical model by using the feature set and the machine learning method;

S06: using the statistical model to match programs in the electronic program table database;

S07: Output the matching result to the user and complete the program recommendation.

In this embodiment, the user inputs the program of interest or the space of interest of the user, and the input content may be a keyword or a phrase or sentence describing the user's preference.

In this embodiment, S02 may specifically include:

S02a: when the received language information is a keyword set, logically calculating the keyword set, and extracting relevant electronic program information from the electronic program table database; and/or,

S02b: When the received language information is a phrase or a sentence, the word segmentation process is first performed, the word segmentation result is calculated to obtain a user-preferred space model, and then the similarity between the space model and the electronic program information in the electronic program table database is calculated, according to the similarity Extract relevant electronic program information.

Specifically, in S02a, a program can be directly extracted from the EPG database by using a keyword set, and each keyword in the keyword set can be implemented by using logical operations such as logical AND, logical OR, logical NOT, logical and non-operation. Specifically, in S02b, the word segmentation tool can be used for processing, and then the user's favorite space model can be obtained by using the method of calculating the word frequency for the word segmentation result, and then the similarity between the space model and the electronic program information in the EPG database is calculated, and then the recommendation result is sorted.

In addition, in S02, the extracted electronic program information may be provided to the user, and the user may perform initial screening, and then the result of the user screening confirmation is used as the extracted electronic program information.

The EPG database of the embodiment of the present invention may be composed of an electronic program table according to a certain structure or a semi-structured structure, which is the same as that described in Embodiment 1, and details are not described herein again. The knowledge base involved in this embodiment includes synonymous, synonymous, antisense, similarity of concepts, and any one or more of words, word classes and semantic attributes, which are specifically described in Embodiment 1. , will not repeat them here.

In this embodiment, there are various methods for performing feature selection in S03, including but not limited to: feature extraction method based on document frequency, information gain method, X ² statistical method, mutual information method, and the like. The feature selection may be performed based on the feature weights, and the feature weights are also calculated in many ways, such as a Boolean weight, an absolute word frequency TF, an IDF, a TF-IDF, a TFC, an ITC, an entropy weight, a TF-IWF, etc., in the embodiment of the present invention. This is not specifically limited. The association information of the feature elements obtained from the knowledge base in S04 includes: attribute information such as semantics and concepts of the words, and the information can be used as a feature set of the user's interest and favorite space, thereby providing data conditions and judgment basis for modeling.

In this embodiment, the machine learning methods used in S05 are various, such as supervised machine learning methods or unsupervised machine learning methods, and semi-supervised machine learning methods; in particular, such as support vector machine (SVM), decision making The decision tree is implemented by any one of the algorithms such as the Bayesian, the maximum entropy, and the conditional random field. It can also be implemented by using a plurality of construction hybrid algorithms, which are not specifically limited in the embodiment of the present invention.

In this embodiment, the matching result may be further sorted in S06, and correspondingly, the sorted result is output to the user in S07.

In this embodiment, the result of the S07 output program recommendation can be used by the user in various forms, such as file output, display output, etc. When there are multiple output results, the screen can be displayed to the user on one screen, or The embodiment of the present invention is not specifically limited. After obtaining the recommendation result, the user can request to play the program that he or she needs, and receive the corresponding data stream for viewing.

In this embodiment, the clustering or classification algorithm may be used in the S03 before or after the feature selection, and the clustering or classification algorithm may be used in the S05 before or after the statistical model is constructed, thereby further improving the program recommendation. Accuracy, which is not specifically limited in this embodiment of the present invention.

The above method provided by the embodiment extracts related electronic program information from the electronic program table database according to the language information input by the user, and performs feature selection to obtain feature elements, and invokes information stored in the knowledge base to enter the feature element. The line is expanded to obtain the feature set of the user's interest space, and the feature set and the machine learning method are used to construct the statistical model, thereby matching the electronic program table database to output the matching result to the user, realizing the program recommendation, and solving the "cold" of the prior art. Start the "problem, and improve the accuracy, performance and usability of the program recommendation. Since the foregoing method is performed on the user side, the user information of the user is not involved in the server side or the user end of the network side. Therefore, the privacy information of the user can be fully protected from leakage and the confidentiality is improved. In addition, the electronic program table database can be searched and evaluated by using the feature element as a search keyword, and then the program pre-selection is performed again according to the processing result, thereby further expanding the user's interest and interest area and improving the accuracy of the program recommendation. Example 4

On the basis of Embodiment 3, this embodiment provides a program recommendation method, which is improved in that the electronic program information is extracted from the EPG database again according to the obtained feature elements, thereby constructing a feature set, as shown in FIG. The method specifically includes:

S11: receiving language information input by the user;

S12: extract relevant electronic program information from an electronic program table database that has stored electronic program information according to the language information;

S13: performing feature selection on the extracted electronic program information to obtain a feature element;

S14: searching and evaluating the electronic program information in the electronic program table database by using the feature element as a search keyword, and extracting relevant electronic program information from the electronic program table database according to the processing result;

S15: performing feature selection on the electronic program information extracted according to the language information in S12 and the electronic program information extracted according to the feature element in S14, to obtain a new feature element;

S16: Obtain association information of the new feature element from a knowledge base in which the language knowledge is stored, and construct a feature set;

S17: constructing a statistical model by using the feature set and the machine learning method;

S18: using the statistical model to match programs in the electronic program table database;

S19: Output the matching result to the user, and complete the program recommendation.

Further, in S14, it may be determined whether it is necessary to pre-select, if yes, proceed to S14 and subsequent steps; otherwise, directly select feature information of the electronic program information extracted according to the language information to obtain a feature element, and from the stored language knowledge Obtain the association information of the feature element in the knowledge base, construct the feature set, and then continue to perform the subsequent steps such as S17.

Wherein, whether the above judgment needs to be pre-selected can be performed in various manners, for example, a simple question window can be preset, whether the user needs to re-select the television program, or simultaneously output the feature element as a dynamic similar semantic network diagram. The manner of the present invention is provided to the user for the user to observe and analyze, and the like, which is not specifically limited in the embodiment of the present invention.

The above method provided by the embodiment extracts relevant electronic program information from the electronic program table database according to the language information input by the user, and performs feature selection to obtain feature elements, and calls the information stored in the knowledge base to expand the feature element to obtain the user. A feature set of the hobby space, using the feature set and the machine learning method to construct a statistical model, thereby matching the electronic program table database output matching result to the user, realizing the program recommendation, and solving the "cold start" problem of the prior art, It also improves the accuracy, performance and usability of program recommendations. Since the above method is performed on the user side, it does not involve collecting user personal information on the network server side or the user end. Therefore, the privacy information of the user can be fully protected from leakage and the confidentiality is improved. In addition, the electronic program table database can be searched and evaluated by using the feature element as a search keyword, and then the program pre-selection is performed again according to the processing result, thereby further expanding the user's interest and interest area and improving the accuracy of the program recommendation. All or part of the above technical solutions provided by the embodiments of the present invention may be completed by hardware related to program instructions, and the program may be stored in a readable storage medium, and the storage medium includes: a ROM, a RAM, a magnetic disk or an optical disk. And other media that can store program code.

The above is only the preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., which are within the spirit and scope of the present invention, should be included in the protection of the present invention. Within the scope.

Claims

Claim

A program recommendation device, wherein the device comprises:

An input unit, configured to receive language information input by a user;

2. The apparatus according to claim 1, wherein the program preselecting unit comprises:

The device according to claim 1, wherein the program preselecting unit comprises:

The device of claim 1, wherein the feature extraction unit further comprises:

a feedback subunit, configured to perform retrieval and evaluation processing on the electronic program information in the electronic program table database by using the feature element as a retrieval keyword, and feed back a result of the processing to the program preselecting unit;

5. Apparatus according to any one of claims 1 to 4 wherein said knowledge base comprises words Synonymous, synonymous, antisense, conceptual similarity, and any one or more of words, word classes, and semantic attributes.

6. A program recommendation method, the method comprising:

Receiving language information input by the user;

The result of the matching is output to the user.

The method according to claim 6, wherein the extracting the related electronic program information from the electronic program table database in which the electronic program information has been stored according to the language information comprises:

8. The method according to claim 6, wherein extracting related electronic program information from the electronic program table database in which the electronic program information has been stored according to the language information comprises:

The method of claim 6, wherein the extracting the electronic program information to perform feature selection, and obtaining the feature element, further comprising:

Correspondingly, acquiring association information of the feature element from a knowledge base in which the language knowledge is stored, and constructing a feature set, include:

The association information of the new feature element is obtained from the knowledge base in which the language knowledge is stored, and the feature set is constructed.

The method according to any one of claims 6 to 9, wherein the knowledge base comprises synonymous, synonymous, antisense, concept similarity, and word, word class and semantic attribute of a word. Any one or more of them.