WO2019198950A1 - Apparatus for providing content information and method therefor - Google Patents

Abstract

An apparatus for providing content information, according to an embodiment of the present invention, may comprise: a first processor for generating management information of viewed contents using a simultaneous generation matrix on the basis of a simultaneous viewing frequency of the viewed contents, and providing similar content information of the viewed contents on the basis of the management information; and a second processor for updating the simultaneous generation matrix on the basis of the simultaneous viewing frequency of the viewed contents on the basis of similar characteristic information when there is no management information of the viewed content.

Description

Apparatus and method for providing content information

The present invention relates to a content information providing apparatus and a method thereof, and more particularly, to a technology for providing content information of similar content when viewing content provided through the Internet.

With the development of the Internet, an environment is created in which not only companies but also individuals can conveniently create and provide contents, and various types and large amounts of contents are provided to users through Internet services.

Among these vast amounts of contents, a search technology for selecting and providing content desired by a user is being developed. In addition to the search technology, similar content is provided by providing the user with content information of similar content related to the content desired by the user. The service to induce the viewing of is also provided.

Conventionally, similar content information is extracted and provided based on similar property information of contents, or contents viewed simultaneously are determined to have high similarity, and thus content with high simultaneous viewing frequency is provided as similar content information. However, when similar content is provided based on the simultaneous viewing frequency, there is a problem in that it is difficult to provide similar content information on the newly registered content because there is no information on the simultaneous viewing frequency of the newly registered content.

Embodiments of the present invention provide a content information providing apparatus and method for providing similar content information as recommended content to a user.

Technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the following descriptions.

An apparatus for providing content information according to an embodiment of the present invention generates management information of the viewed contents by using a co-occurrence matrix based on the simultaneous viewing frequency of the viewed contents, and based on the management information, the similarity of the viewed contents. A first processor providing content information; And a second processor for updating the simultaneous generation matrix based on the simultaneous viewing frequency of the viewed content based on the similar characteristic information when there is no management information of the viewed content.

In an embodiment, the first processor calculates a correlation between the watched contents based on the co-occurrence matrix, classifies the watched contents on a vector space based on the correlation, and manages the management information. May include generating.

In one embodiment, the management information may include location information of the viewed contents on a vector space.

In one embodiment, the co-occurrence matrix may include a simultaneous viewing frequency of two contents forming a data pair.

In one embodiment, the second processor, the similar content information of the content that does not exist the management information by using an algorithm learned based on the similar property information does not exist the watched content and the management information. Calculating a similarity probability value between the contents that are not included.

In an embodiment, the second processor may include calculating a hyperparameter based on a frequency of simultaneous occurrence of content having the highest similarity probability value.

In an embodiment, the second processor may include calculating, as the hyperparameter, the largest number of total simultaneous occurrence frequencies or simultaneous occurrence frequencies of content having the highest similarity probability value.

In an embodiment, the second processor may add the content in which the management information does not exist in the row and column of the co-occurrence matrix as an item, calculate the similarity probability value and the hyperparameter of the watched content, respectively, and manage the management. And calculating the co-occurrence frequency of content for which there is no information to update the co-occurrence matrix.

In an embodiment, the second processor may include calculating a correlation between the entire contents including the content in which the management information does not exist based on the updated co-occurrence matrix.

In an embodiment, the second processor calculates a correlation between the entire contents based on a co-occurrence matrix that does not include the content in which the management information does not exist, and the management information exists based on the similarity probability value. And calculating correlation between content not to be viewed and content viewed in the co-occurrence matrix.

In an embodiment, the first processor may include generating management information of content in which the management information does not exist based on a correlation calculated by the second processor, and providing the similar content information. have.

According to an embodiment of the present invention, a method of providing content information generates management information of the viewed contents using a co-occurrence matrix based on the simultaneous viewing frequency of the viewed contents, and generates the viewed information based on the management information. Providing similar content information; And updating the simultaneous generation matrix based on the simultaneous viewing frequency of the viewed content based on the similar characteristic information when there is no management information of the viewed content.

In an embodiment, the method may further include providing similar content information of content in which the management information does not exist based on the updated co-occurrence matrix.

In an embodiment, the providing of the similar content information may include calculating a correlation between the viewed contents based on the co-occurrence matrix, and classifying the viewed contents on a vector space based on the correlation. Generating management information.

The updating of the co-occurrence matrix may include: managing the watched content and the management using similar property information of content in which the management information does not exist using an algorithm learned based on similar property information. The method may include calculating a similarity probability value between contents in which information does not exist.

In an embodiment, updating the co-occurrence matrix may include calculating a hyperparameter based on a co-occurrence frequency of content having the highest similarity probability value.

In an embodiment, the updating of the co-occurrence matrix may include calculating, as the hyperparameters, the total sum of all co-occurrence frequencies of the content having the highest similarity probability value or the co-occurrence frequency as the hyperparameter, Computing each of the hyperparameters may be calculated as the frequency of the simultaneous occurrence of the content that does not exist the management information to update the co-occurrence matrix.

In an embodiment of the present disclosure, the providing of the similar content information may include calculating a correlation between the entire contents including the content without the management information based on the updated co-occurrence matrix.

In one embodiment, the providing of the similar content information may include: calculating a correlation between all contents based on a co-occurrence matrix that does not include content in which the management information does not exist, and based on the similarity probability value. The method may include calculating a correlation between content for which no management information exists and watched contents included in the co-occurrence matrix.

In an embodiment, the providing of the similar content information may include determining the content of the content in which the management information does not exist based on a correlation between the content in which the management information does not exist and the viewed content included in the co-occurrence matrix. Generating management information and providing the similar content information.

According to an apparatus and method for providing content information according to embodiments of the present invention, similar content information may be provided to a user as recommended content.

According to an apparatus and method for providing content information according to embodiments of the present invention, the accuracy of similar content information provided as recommended content to a user may be improved.

In addition, various effects may be provided that are directly or indirectly identified through this document.

1 is a block diagram illustrating a configuration of an apparatus for providing content information according to an embodiment of the present invention.

2 is a diagram illustrating an example of a data pair of simultaneously viewed contents for generating a co-occurrence matrix according to an embodiment of the present invention.

3 is an exemplary diagram for generating a co-occurrence matrix using the data pair of FIG. 2.

4A and 4B are diagrams for describing an example of learning a relative position of a co-occurrence matrix-based contents in a vector space according to an embodiment of the present invention.

FIG. 5 is a diagram illustrating a method of learning similarity of contents using a neural network trained based on a co-occurrence matrix according to an embodiment of the present invention.

FIG. 6 is a diagram illustrating a method of calculating similarity probability values of existing contents for new contents using a neural network trained based on a co-occurrence matrix according to an embodiment of the present invention.

7 is an exemplary diagram of updating a co-occurrence matrix for new content by calculating a maximum frequency-based hyperparameter according to an embodiment of the present invention.

8 is an exemplary diagram of updating a co-occurrence matrix for new content by calculating a total frequency-based hyperparameter according to an embodiment of the present invention.

9 is an exemplary view illustrating a method of calculating a correlation PMI according to an embodiment of the present invention.

10 is an exemplary diagram of updating a co-occurrence matrix for new content by calculating a correlation (PMI) according to an embodiment of the present invention.

11 is a flowchart illustrating a method of generating and managing a co-occurrence matrix for content according to an embodiment of the present invention.

12 is a flowchart illustrating a content providing method according to an embodiment of the present invention.

FIG. 13 is a flowchart illustrating a method of updating a co-occurrence matrix based on similar property information for new content according to an embodiment of the present invention.

14 is a flowchart illustrating a method of updating a co-occurrence matrix based on similar property information for new content according to another embodiment of the present invention.

15 illustrates a computing system in accordance with an embodiment of the present invention.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible even though they are shown in different drawings. In addition, in describing the embodiments of the present invention, if it is determined that the detailed description of the related well-known configuration or function interferes with the understanding of the embodiments of the present invention, the detailed description thereof will be omitted.

In describing the components of the embodiment of the present invention, terms such as first, second, A, B, (a), and (b) may be used. These terms are only for distinguishing the components from other components, and the nature, order or order of the components are not limited by the terms. In addition, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art, and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

Hereinafter, embodiments of the present invention will be described in detail with reference to FIGS. 1 to 15.

1 is a block diagram illustrating a configuration of an apparatus for providing content information according to an embodiment of the present invention.

Referring to FIG. 1, the content information providing apparatus 100 includes a first processor 110 and a second processor 120. The first processor 110 and the second processor 120 may be implemented as a learned network. In FIG. 1, the first processor 110 and the second processor 120 are shown in separate configurations, but are not limited thereto. The first processor 110 and the second processor 120 may be integrated. It may also operate as a processor.

The learned neural network of the first processor 110 generates a co-occurrence matrix based on the learned algorithm (eg, the first algorithm) and based on the co-occurrence matrix. Management information of contents may be generated. Here, the learned algorithm may be defined as an algorithm for generating similar contents to have similar management information values. For example, contents that are similar to each other may have management information located adjacent to a vector space, and may mean the same or similar contents in content, subject, material, appearance, category, summary, and the like of the content. .

The learned neural network of the second processor 120 updates the co-occurrence matrix by calculating similarity probability values between the contents based on the similar characteristic information of the contents based on the learned algorithm (eg, the second algorithm). Here, the similarity probability value means the degree of similarity between contents, and the calculation of the similarity probability value may be implemented by a general similarity related algorithm, and the learning network of the second processor 120 may be learned by such an algorithm. have. In addition, the 'similar property information' may include the same or similar property information in content, theme, location, appearance, category, summary, and the like of the content. For example, if the content is a movie, it may include a main actor, release year, movie title, production cost and the like.

The first processor 110 may generate management information of the viewed contents using a simultaneous generation matrix based on the simultaneous viewing frequency of the viewed contents, and provide similar content information of the viewed contents based on the management information. In this case, the first processor 110 may generate management information of the contents at predetermined intervals or may generate the management information every predetermined time after the new content is registered.

The first processor 110 generates a co-occurrence matrix using the simultaneous viewing frequency of the contents simultaneously viewed, calculates a pointwise mutual information (PMI) of the contents using the value of the co-occurrence matrix, and correlates the contents. Based on the diagram, contents may be classified to have a relative position in an N-dimensional vector space. A method of generating the co-occurrence matrix will be described in more detail later with reference to FIGS. 2 and 3.

Here, the rows and columns of the co-occurrence matrix may include identification information (eg, content name) of the contents and a frequency at which two contents forming a data pair are simultaneously viewed. At this time, the condition to be viewed at the same time may include the case where the viewing of the content 2 is started while the content 1 is being viewed. In addition, when the content 2 is viewed within a predetermined time after the content 1 is completed viewing, it may be regarded as simultaneous viewing.

The first processor 110 calculates a correlation between the contents using the value of the co-occurrence matrix, and classifies the contents on the N-dimensional vector space based on the correlation. In this case, each of the contents may be arranged in positions adjacent to each other as the frequency of occurrence or correlation is high in the N-dimensional vector space, and the first processor 110 may determine that the similarities of the contents in the adjacent positions are high. Can be. That is, the first processor 110 may classify the contents in the adjacent position as the similar contents and generate management information (vector information) which is relative position information on the vector space for each contents.

When there is no management information of the watched content, the second processor 120 may update the co-occurrence matrix based on the simultaneous viewing frequency of the watched content in which the management information does not exist based on the similar characteristic information of the content. .

The second processor 120 may calculate a similarity probability value between the watched content and the new content by using similar property information of the content (new content) in which the management information does not exist using an algorithm learned based on the similar property information. Can be. For example, the new content may include content that is newly registered and is not included in the co-occurrence matrix, so that management information is not generated. That is, when the characteristic information of the new content is input to the second processor 120, the second processor 120 may generate new content and existing content (content already viewed and management information existing or simultaneously generated based on the neural network on which learning has been completed). The similarity probability value between the contents included in the matrix) is calculated.

The second processor 120 may update the co-occurrence matrix by calculating a hyper-parameter based on the co-occurrence frequency of the content having the highest similarity probability value. The second processor 120 may calculate the total sum of all simultaneous occurrence frequencies of the content having the highest similarity probability value as a hyper parameter. In addition, the second processor 120 may calculate, as a hyper parameter, the largest number of simultaneous occurrence frequencies of content having the highest similarity probability value.

The second processor 120 adds new content to the rows and columns of the co-occurrence matrix as items, calculates similarity probability values and hyperparameters of the watched contents, respectively, and calculates the co-occurrence matrix by updating the co-occurrence matrix. Can be. Here, 'operation' may include a multiplication operation.

The second processor 120 may calculate a correlation between the entire contents based on the updated co-occurrence matrix based on the hyperparameter and the similarity probability value. In this case, a correlation between the new content and the existing content may also be calculated.

In another embodiment, the second processor 120 does not calculate a hyperparameter, but calculates a correlation between the entire contents using a similarity probability value between the new content and the existing content and a co-occurrence matrix that does not include the new content. Can be.

When the second processor 120 calculates a correlation between the new content and the existing content, the first processor 110 generates the new content based on the correlation between the new content and the existing content calculated by the second processor 120. The management information for the content may be generated, and similar content information may be generated and provided by a learned algorithm.

The content information providing apparatus 100 of the present invention having such a configuration provides similar content information based on the simultaneous viewing frequency, but does not have the simultaneous viewing frequency information, that is, the existing content for the new content that does not generate the simultaneous matrix. Based on a neural network trained based on similar property information of contents, a similarity probability value for each of the existing contents is calculated and randomly calculated the simultaneous viewing frequency of the new content based on the similarity probability value. Similar content information may be provided based on the simultaneous viewing frequency.

Hereinafter, a method of generating a co-occurrence matrix will be described with reference to FIGS. 2 and 3.

2 is a diagram illustrating an example of a data pair of simultaneously viewed contents for generating a co-occurrence matrix according to an embodiment of the present invention. 3 is an exemplary diagram for generating a co-occurrence matrix using the data pair of FIG. 2.

In FIGS. 2 and 3, contents are described as examples for convenience of description, but the contents may include various multimedia data for the purpose of shopping and advertisement. The process of generating the co-occurrence matrix of FIGS. 2 and 3 may be implemented by the first processor 110.

Referring to FIG. 2, when movies 1 to 9 are watched, and movies 2 and 3 are sequentially watched while movie 1 is being viewed, the content information providing apparatus 100 may include movie 1, movie 2, and movie 1. And movie 3 are each generated as data pairs.

In addition, when movie 1, movie 3, and movie 4 are sequentially watched while movie 2 is being viewed, the content information providing apparatus 100 includes four data pairs (movie 2 and movie 1, movie 2 and movie 3, movie). 2 and movie 4) are generated.

Referring to FIG. 3, the content information providing apparatus 100 generates a co-occurrence matrix based on the data pairs generated in FIG. 2. That is, the movie 1 and the movie 2 are simultaneously watched twice, the frequency is recorded at 2, and the movie 2 and the movie 3 are also watched twice at the same time, and the frequency is recorded at 2. As such, the content information providing apparatus 100 generates a value of the co-occurrence matrix based on the number of simultaneous viewing.

4A and 4B are diagrams for describing an example of learning a relative position (management information) in a vector space of co-occurrence matrix-based contents according to an embodiment of the present invention.

Referring to FIG. 4A, the first processor 110 is implemented as a one hot vector, and the one hot vector refers to a vector in which only one value is '1' and the rest is '0' in the entire array.

When two words (movie 1 and movie 2), which are data pairs, are input and output, the first processor 110 weights W such that 'movie 1' is input to the input layer and 'movie 2' is output from the output layer. Learn the value of the number of embedding dimensions (N)). When the learning is completed, the relative positions in the vector space of the words (movie 1 and movie 2) are learned to the weight values.

In FIG. 4A, the value of movie 2 is set to '1' as a similar content for movie 1, but the value of '1' may be replaced with a correlation (PMI (movie 1 ';' movie 2 ')) as shown in FIG. 4B. have. That is, the weight values for movie 2, which is highly correlated with movie 1, can be learned. The calculation of the correlation will be described in more detail with reference to FIG. 9.

5 is a diagram illustrating a method of learning similarity of contents using a neural network trained based on a co-occurrence matrix according to an embodiment of the present invention, and FIG. 6 is a co-occurrence according to an embodiment of the present invention. A diagram for describing a method of calculating similarity probability values of existing contents for new contents using a neural network trained on a matrix basis.

5 and 6 disclose an example of calculating similarity or similarity probability values through the learned characteristic network based on the similar characteristic information by the second processor 120. 5 and 6, the neural network is composed of N layers composed of a plurality of units, and the calculated value Y calculated through the layers calculates the similarity or similarity probability value through a softmax function. can do.

Referring to FIG. 5, when similar property information of content is input to a neural network, the second processor 120 outputs a calculated value Y and performs a softmax function to output a similarity degree.

Referring to FIG. 5, when similar characteristic information of movie 2 is input to a neural network learned based on management information of contents (movies 1 to 8), movie 3 located at a location adjacent to movie 2 is based on management information. The weight (w) and the variable (b) are optimized to output a 1 'value. In this case, the similar characteristic information may include a director, a lead actor, a release year, a movie title, a production cost, etc. of the movie 2, and the similar characteristic information of the movie 2 and the movie 3 may coincide.

Referring to FIG. 6, when similar characteristic information of Movie 9, which is new content in which no management information exists, is input to a neural network trained to have the optimized weight w and the variable b, the second processor 120 is input. Calculates the similarity probability using the optimized weight w and the variable b. It can be seen that the similarity probability value of Movie 3 is the highest.

7 is an exemplary diagram of updating a co-occurrence matrix for new content for which management information does not exist by calculating a maximum frequency-based hyperparameter according to an embodiment of the present invention. FIG. 7 illustrates an example of generating a value of a co-occurrence matrix for a new content 'movie 9' by calculating a hyper parameter.

The second processor 120 may calculate the maximum frequency of the content having the highest similarity probability value as the hyper parameter among the contents constituting the co-occurrence matrix, and multiply the hyper parameter by the similarity probability value of each content to calculate the frequency for the new content. have.

In FIG. 7, since the similarity probability value of 'movie 3' is 0.5, the maximum frequency of movie 3 is calculated as a hyper parameter. That is, the simultaneous viewing frequency of movie 3 with movie 1 is 2, the simultaneous viewing frequency of movie 3 and movie 2 is 2, the simultaneous viewing frequency of movie 3 and movie 4 is 2, and the simultaneous viewing frequency of movie 3 and movie 5 is 1 Therefore, the content information providing apparatus 100 calculates the maximum frequency '2' as a hyper parameter. Subsequently, the second processor 120 may calculate a frequency for 'movie 9' by multiplying the hyperparameter '2' by the similarity probability value of each content.

In this case, the frequency of movie 9 and movie 1 is 0.4, the frequency of movie 9 and movie 2 is 0.4, the frequency of movie 9 and movie 3 is 1.0, the frequency of movie 9 and movie 4 is 0.4, and the frequency of movie 9 and movie 5 is Since 0.2, the second processor 120 may record only the frequency of the movie 3 having the largest value as '1' and the remaining contents as '0'. In this case, the second processor 120 may record values obtained by multiplying the hyperparameter and the similarity probability value for each content corresponding to the new content having no management information. In this case, the number recorded may be discarded or rounded down to a decimal point to record only an integer.

8 is an exemplary diagram of updating a co-occurrence matrix for new content by calculating a total frequency-based hyperparameter according to an embodiment of the present invention. 8 illustrates an example of generating a value of a co-occurrence matrix for the new content "movie 9" by calculating a hyper parameter.

 The second processor 120 adds the frequency of the content having the highest likelihood probability value among the contents constituting the co-occurrence matrix, calculates the sum as a hyperparameter, multiplies the hyperparameter with the similarity probability value of each content, and generates a frequency for the new content. Can be calculated.

In FIG. 8, since the similarity probability value of 'movie 3' is the largest as 0.5, the sum of the frequency of movie 3 is calculated as a hyper parameter. That is, the simultaneous viewing frequency of movie 3 with movie 1 is 2, the simultaneous viewing frequency of movie 3 and movie 2 is 2, the simultaneous viewing frequency of movie 3 and movie 4 is 2, and the simultaneous viewing frequency of movie 3 and movie 5 is 1 Therefore, the second processor 120 calculates a hyper parameter '7', which is the sum of the frequencies.

Subsequently, the second processor 120 may calculate the frequency for the movie 9 by multiplying the hyperparameter '7' by the similarity probability value of each content.

In this case, the frequency of movie 9 and movie 1 is 1.4, the frequency of movie 9 and movie 2 is 1.4, the frequency of movie 9 and movie 3 is 3.5, the frequency of movie 9 and movie 4 is 1.4, and the frequency of movie 9 and movie 5 is Since 0.7, the second processor 120 may record only the frequency of the movie 3 having the largest value as '1' and the remaining contents as '0'. In this case, the second processor 120 may record all of the content obtained by multiplying the hyperparameter and the similarity probability value for each content, but may discard only the decimal point and record only the integer.

FIG. 9 illustrates an example of calculating Pointwise Mutual Information (PMI) between contents according to an embodiment of the present invention, and FIG. 10 simultaneously includes information on new content according to an embodiment of the present invention. It is a figure which shows an example of an generation matrix.

The first processor 110 calculates a correlation between the contents of the co-occurrence matrix to generate management information between the contents, and the second processor 120 correlates the contents of the co-occurrence matrix updated based on the hyperparameter. The degree of correlation may be calculated between the contents of the existing co-occurrence matrix before the update or the update, and the correlation between the new content and the existing contents may be calculated based on the similarity probability value of the new content. The method is described in detail through the following equations.

Correlation between the contents x and y may be defined as in Equation 1 below.

p (x) is the probability that x content is watched, p (y) is the probability that y content is watched, p (x, y) is the probability that x content and y content are watched simultaneously.

Equation 1 may be expressed as Equation 2.

x _i is the sum of the values of the rows of the content with the highest similarity probability value, x _j is the sum of the values of the columns of the content with the highest similarity probability value, and D is the sum of the frequencies of the co-occurrence matrix (see FIG. 9).

The first processor 110 and the second processor 120 may calculate a correlation between existing contents of the co-occurrence matrix through Equation 2. However, in order to calculate the correlation between the new content and the existing content, the second processor 110 assumes a similarity probability value as shown in Equation 3 below.

When the equation is represented by substituting the similarity probability value of Equation 3 into Equation 2, Equation 4 below.

The second processor 110 may calculate the correlation PMI of the new content by using the similarity probability value as shown in Equation 4.

For example, in order to calculate the correlation between the existing contents of the movie 2 and the movie 3, the equation 2 is applied to calculate the equation.

The second processor 120 inputs and learns the original hot vector as shown in FIG. 4B based on the correlation between the contents. That is, the second processor 120 optimizes the weight so that the movie 3 is output when the movie 2 is input to the neural network and the movie 3 is set as the output. Then, when the learning is completed, the second processor 120 learns the relative position in the vector space of the movie 2 and the movie 3 by using the weight values. That is, the correlation between the probability of watching each movie may be learned through the weight value. That is, the correlation between the movies 1 to 8 of FIG. 10 may be calculated through Equation 5.

On the other hand, when calculating the correlation between the new content and the existing content Equation 4 is applied. Referring to FIG. 10, Equation 2 described above is applied to the correlation between Movies 1 to 8, and Equation 4 is applied to the calculation of the correlation between the new content (Movie 9) and the existing content (Movies 1 to 8).

Table 1 below shows an example of calculating correlation between movie 9 and existing contents (movies 1 to 8) by applying x _j and D values of FIG. 9 to Equation 4, and table 2 displays final calculated results. One example table.

Movie 1 Movie 2 Movie 3 Movie 4 Movie 5 Movie 6 Movie 7 Movie 8 Movie 9 Movie 9

0 0 0 0

Movie 1 Movie 2 Movie 3 Movie 4 Movie 5 Movie 6 Movie 7 Movie 8 Movie 9 Movie 9 0.1139 -0.0621 0.2688 -0.0621 0.1139 0 0 0 -

In Table 2, it can be seen that movie 9 has the highest correlation with movie 3.

The second processor 120 may calculate all the correlations between the movies 1 to 9 and display them as shown in FIG. 10, and the first processor 110 manages the respective contents based on the calculated correlations between the entire contents. When the movie 9 is watched by generating the information, the movie 3 may be recommended as similar content information.

11 is a flowchart illustrating a method of generating and managing a co-occurrence matrix for content according to an embodiment of the present invention. Hereinafter, it is assumed that the content information providing apparatus 100 of FIG. 1 performs the process of FIG. 11. In addition, in the description of FIG. 11, it can be understood that the operations described as being performed by the apparatus are controlled by the first processor 110.

Referring to FIG. 11, when content viewing starts (S110), when there is at least one or more contents simultaneously viewed with the content being viewed (S120), the content viewing apparatus 100 simultaneously views the simultaneously viewed contents. The frequency is updated in the co-occurrence matrix (S130).

The content information providing apparatus 100 calculates a correlation PMI between contents based on the updated co-occurrence matrix (S140). Subsequently, the content information providing apparatus 100 classifies the contents on the vector space using the correlation to generate management information (vector information) (S150). In this case, the vector information is position information in a vector space, and the positions of contents having high similarity are adjacent to each other. In addition, the content information providing apparatus 100 may calculate the correlation between the contents through Equation 2 described above.

The content information providing apparatus 100 provides similar content information of the viewed content based on the generated management information (S160).

Hereinafter, a content providing method according to an exemplary embodiment of the present invention will be described in detail with reference to FIG. 12. Hereinafter, it is assumed that the content information providing apparatus 100 of FIG. 1 performs the process of FIG. 12.

When the content is viewed (S210), the content information providing apparatus 100 checks whether there is concurrent matrix information on the viewed content (S220).

If there is simultaneous generation matrix information on the viewed content, the content information providing apparatus 100 provides similar content information of the viewed content on the basis of the simultaneous generation matrix (S230).

On the other hand, if there is no co-occurrence matrix information for the viewed content, the content information providing apparatus 100 updates the co-occurrence matrix by using the neural network learned based on the similar characteristic information. Thereafter, the apparatus 100 for providing content information provides similar content information based on the updated co-occurrence matrix (S240). That is, the management information of each content is generated using the updated co-occurrence matrix, and similar content information is provided based on the management information.

12 illustrates an example in which the content information providing apparatus 100 updates a co-occurrence matrix based on similar property information when there is no management information of the viewed content, but the present invention is not limited thereto. The information providing apparatus 100 may be implemented to periodically update the co-occurrence matrix based on the similar characteristic information.

Hereinafter, a method of updating a co-occurrence matrix based on similar property information for new content according to an embodiment of the present invention will be described with reference to FIG. 13. FIG. 13 is a flowchart illustrating a method of updating a co-occurrence matrix based on similar property information with respect to new content according to an embodiment of the present invention. FIG. A method of updating the co-occurrence matrix for Hereinafter, it is assumed that the content information providing apparatus 100 of FIG. 1 performs the process of FIG. 13. In addition, in the description of FIG. 13, it can be understood that S310 to S340 are controlled by the second processor 120 and the process of S350 is controlled by the first processor 110 among the operations described as being performed by the apparatus. have.

The content information providing apparatus 100 calculates each similarity probability value for contents that are items of a co-occurrence matrix. That is, the content information providing apparatus 100 calculates a similarity probability value using a neural network that has been previously learned based on similar characteristic information of new content (S310). In this case, referring to FIG. 6, the content information providing apparatus 100 may calculate similarity probability values of respective contents by inputting similar characteristic information of the new movie 9 into the neural network (second processor 120) that has already been learned. Can be.

The content information providing apparatus 100 calculates a hyper parameter based on the simultaneous viewing frequency of the contents which are the concurrent occurrence matrix items (S320). In this case, the hyperparameter may be calculated from the sum or the maximum frequency of the frequencies of the content having the highest similarity probability value. In this case, as shown in FIGS. 7 and 8, the hyper parameter may be calculated as the sum value or the maximum value of the row.

The content information providing apparatus 100 updates the co-occurrence matrix for the new content with a value obtained by multiplying the similarity probability value and the hyperparameter (S330). The content information providing apparatus 100 may store a value obtained by multiplying a hyperparameter with a similarity probability value of each of the contents (movies 1 to 8) for the new content, movie 9, as the frequency for the movie 9 item in the co-occurrence matrix. have.

The content information providing apparatus 100 calculates a correlation PMI of all contents (new content and existing content) based on the updated co-occurrence matrix (S340). When the co-occurrence matrix is updated through the hyperparameter as shown in FIG. 7 or FIG. 8, the content information providing apparatus 100 may calculate a correlation between the entire contents (movies 1 to 9) through Equation 2.

Thereafter, the content information providing apparatus 100 inputs a correlation degree (PMI) between the new content and the existing content to the first processor 110 to generate management information of the new content and generate similar content information based on the management information. Provided (S350). The first processor 110 may generate a weight value obtained by inputting a correlation to the neural network as shown in FIG. 4B as management information of contents.

As described above, FIG. 13 discloses an example in which similar content information is provided by calculating a correlation between all contents (including new content) after updating a co-occurrence matrix for new content based on a hyper parameter.

Hereinafter, a method of updating a co-occurrence matrix based on similar property information for new content according to another embodiment of the present invention will be described with reference to FIG. 14. FIG. 14 is a flowchart illustrating a method of updating a co-occurrence matrix based on similar property information with respect to new content according to another embodiment of the present invention. FIG. 12 is new content without management information of process S240 of FIG. 12. Corresponding to another embodiment specifically illustrating a method of updating a co-occurrence matrix for.

Hereinafter, it is assumed that the content information providing apparatus 100 of FIG. 1 performs the process of FIG. 14. In addition, in the description of FIG. 14, S410 to S420 among the operations described as being performed by the apparatus may be understood to be controlled by the second processor 120 and the process of S430 may be controlled by the first processor 110. have.

The content information providing apparatus 100 calculates each similarity probability value for contents that are items of a co-occurrence matrix. That is, the content information providing apparatus 100 calculates a similarity probability value using a neural network that has been previously learned based on similar characteristic information of new content (S410). In this case, referring to FIG. 6, the content information providing apparatus 100 may calculate similarity probability values of respective contents by inputting similar characteristic information of movie 9, which is new content, to a neural network (second processor 120) that has already been learned. Can be.

Subsequently, the content information providing apparatus 100 calculates a correlation between the entire contents based on the similarity probability value and the co-occurrence matrix that does not include the new contents (S420). That is, the content information providing apparatus 100 calculates a correlation between movies 1 to 8 by applying Equation 2 based on a co-occurrence matrix that does not include new content, and between the new content (movie 9) and existing contents. The correlation between the new content (movie 9) and the existing content may be calculated by applying Equation 4 based on the similarity probability value and the concurrent matrix including no new content.

Correlation PMI (Movie 1, Movie 2) to PMI (Movie 7, Movie 8) is calculated through Equation 2 with respect to Movies 1 through 8 in FIG. 10, and Correlation PMI through Equation 4 with respect to Movie 9. (Movie 9, Movie 1) to PMI (Movie 8, Movie 9) are calculated.

Thereafter, the content information providing apparatus 100 inputs a correlation degree (PMI) between the new content and the existing content to the first processor 110 to generate management information of the new content and generate similar content information based on the management information. Provided (S430). The first processor 110 may generate a weight value obtained by inputting a correlation to the neural network as shown in FIG. 4B as management information of contents.

As described above, the apparatus 100 for providing content information of the present invention generates a co-occurrence matrix, generates management information of respective contents based on the generated co-occurrence matrix, and provides similar content information based on the management information. In addition, the content information providing apparatus 100 updates the co-occurrence matrix for the new content by using the learned neural network based on the similar characteristic information of the new content without the management information.

15 illustrates a computing system in accordance with an embodiment of the present invention.

Referring to FIG. 15, the computing system 1000 may include at least one processor 1100, a memory 1300, a user interface input device 1400, a user interface output device 1500, and storage connected through a bus 1200. 1600, and network interface 1700.

The processor 1100 may be a central processing unit (CPU) or a semiconductor device that executes processing for instructions stored in the memory 1300 and / or the storage 1600. The memory 1300 and the storage 1600 may include various types of volatile or nonvolatile storage media. For example, the memory 1300 may include a read only memory (ROM) and a random access memory (RAM).

Thus, the steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, software module, or a combination of the two executed by the processor 1100. The software module resides in a storage medium (ie, memory 1300 and / or storage 1600), such as RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM. You may.

An exemplary storage medium is coupled to the processor 1100, which can read information from and write information to the storage medium. In the alternative, the storage medium may be integral to the processor 1100. The processor and the storage medium may reside in an application specific integrated circuit (ASIC). The ASIC may reside in a user terminal. In the alternative, the processor and the storage medium may reside as discrete components in a user terminal.

The above description is merely illustrative of the technical idea of the present invention, and those skilled in the art to which the present invention pertains may make various modifications and changes without departing from the essential characteristics of the present invention.

Therefore, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention but to describe the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

As described above, according to the apparatus and method for providing content information according to embodiments of the present invention, similar content information may be provided to the user as recommended content, and the accuracy of the similar content information provided as the recommended content to the user may be improved. Can be improved.

In addition, a preferred embodiment of the present invention is for the purpose of illustration, those skilled in the art will be able to various modifications, changes, replacements and additions through the spirit and scope of the appended claims, such modifications and changes are claimed in the following claims It should be seen as belonging to a range.

Claims (20)

A first processor generating management information of the viewed contents using a simultaneous generation frequency based on the simultaneous viewing frequency of the viewed contents and providing similar content information of the viewed contents based on the management information; A second processor for updating a co-occurrence matrix based on the co-view frequency of the watched content based on the similar characteristic information when there is no management information of the watched content Content information providing apparatus comprising a. The method according to claim 1, The first processor, And calculating the correlation between the viewed contents based on the co-occurrence matrix, and classifying the viewed contents into a vector space based on the correlation to generate management information. The method according to claim 2, The management information, Content information providing apparatus comprising the position information of the watched content on a vector space. The method according to claim 1, The co-occurrence matrix, And content viewing frequency of two contents forming a data pair simultaneously. The method according to claim 1, The second processor, A similarity probability value is calculated between the watched content and the content without the management information by using the similar property information of the content without the management information by using an algorithm learned based on the similar property information. Content information providing apparatus. The method according to claim 5, The second processor, And calculating a hyperparameter based on a frequency of simultaneous occurrence of the content having the highest similarity probability value. The method according to claim 6, The second processor, And calculating, as the hyperparameter, the largest of the total sum of the simultaneous occurrence frequencies or the simultaneous occurrence frequencies of the content having the highest similarity probability value as the hyper parameter. The method according to claim 6, The second processor, Add the content in which the management information does not exist in the row and column of the co-occurrence matrix as an item, And calculating the similarity probability value and the hyperparameter of the viewed contents, respectively, as a frequency of simultaneous occurrence of the content in which the management information does not exist, and updating the simultaneous occurrence matrix. The method according to claim 8, The second processor, And a correlation between all contents including contents without the management information based on the updated co-occurrence matrix. The method according to claim 5, The second processor, Calculating a correlation between the entire contents based on a co-occurrence matrix that does not include the contents in which the management information does not exist, And a correlation between the content in which the management information does not exist and the viewed content included in the co-occurrence matrix is calculated based on the similarity probability value. The method of claim 9, wherein the first processor generates management information of content in which the management information does not exist based on a correlation calculated by the second processor, and provides the similar content information. Content information providing apparatus. Generating management information of the viewed contents using a co-occurrence matrix based on the simultaneous viewing frequency of the viewed contents, and providing similar content information of the viewed contents based on the management information; And If there is no management information of the watched content, updating the co-occurrence matrix based on the co-viewing frequency of the watched content based on the similar characteristic information; Content information providing method comprising a. The method according to claim 12, Providing similar content information of content in which the management information does not exist based on the updated co-occurrence matrix; The content information providing method further comprising. The method according to claim 12, Providing the similar content information And calculating correlation between the watched contents based on the co-occurrence matrix, and classifying the watched contents into a vector space based on the correlation to generate management information. The method according to claim 13, Updating the co-occurrence matrix, A similarity probability value is calculated between the watched content and the content without the management information by using the similar property information of the content without the management information by using an algorithm learned based on the similar property information. How to provide content information. The method according to claim 15, Updating the co-occurrence matrix, And calculating a hyperparameter based on a frequency of simultaneous occurrence of the content having the highest similarity probability value. The method according to claim 16, Updating the co-occurrence matrix, The total sum of all simultaneous occurrence frequencies of the content having the highest similarity probability value or the largest number of simultaneous occurrence frequencies is calculated as the hyper parameter, and the similarity probability value and the hyper parameter are respectively calculated to calculate the content of the content that does not exist. The content information providing method, characterized by updating the co-occurrence matrix by calculating the co-occurrence frequency. The method according to claim 17, Providing the similar content information, And calculating a correlation between the entire contents including the content in which the management information does not exist based on the updated co-occurrence matrix. The method according to claim 13, Providing the similar content information, Calculating a correlation between the entire contents based on a co-occurrence matrix that does not include the contents in which the management information does not exist, And calculating a correlation between the content in which the management information does not exist and the viewed content included in the co-occurrence matrix based on the similarity probability value. The method according to claim 18 or 19, Providing the similar content information, Generating management information of the content in which the management information does not exist and providing the similar content information based on a correlation between the content in which the management information does not exist and the viewed content included in the co-occurrence matrix; Content information providing method.

Images