Patent Application
20130283303
This application claims the benefit under 35 U.S.C. §119(a) of Korean Patent Application No. 10-2012-0042255, filed on Apr. 23, 2012, the entire disclosure of which is incorporated herein by reference for all purposes.
1. Field
The following description relates to a content providing apparatus and method, such as an Internet protocol (IP) TV platform, and more particularly, to a method for searching for and recommending content.
2. Description of the Related Art
Generally, content search and recommendation is conducted based on metadata of content. A user inputs content-related search information, such as the name of content, a main character, cast, and the genre of the content, to a search engine provided by a content providing platform or the Internet to request content recommendations.
In this case, recommendation results may be limited to information that only relates to the metadata of the content that is based on information that the user comes up with. However, the metadata of content is not adequate as search information. For example, when a service user feels depressed, the user may want movies to change their current mood regardless of the usual preference of film genre or actors or actresses. In other words, the user may want films that make the user laugh or cry, giving an emotional catharsis, or the user may not be sure about the exact genre of film that they want to watch.
When being in a particular emotional state, a user may want different content, apart from the usual preference, and in this case, the user may have difficulties finding desired content if only based on the user's existing knowledge. Hence, there may be a content search and recommendation method required for providing content recommendations suitable to a search term indicating a user's emotion, such as depression.
In this regard, there are music selection devices to recommend music based on a user's emotional state. These devices convert user's emotions into numeric data, and recommend music of a specific genre based on the converted numeric data (additionally, contextual information such as time and age). However, unlike music, movies include various forms of media data, and thus there may be a low relevance between their genres and the user's emotions. As described above, some users may prefer comedy movies but others may want sad movies when they feel depressed.
Thus, there is a need of an emotion-based recommendation method for content such as movies, which ensures a high level of satisfaction.
The following description relates to an apparatus and method for recommending multimedia content such as movies based on an emotion keyword.
In addition, the following description relates to an apparatus and method for recommending content based on a user's emotion, by using a database storing acquired emotion information of the user with respect to each content.
In one general aspect,
Other features and aspects will be apparent from the following detailed description, the drawings, and the claims.
Throughout the drawings and the detailed description, unless otherwise described, the same drawing reference numerals will be understood to refer to the same elements, features, and structures. The relative size and depiction of these elements may be exaggerated for clarity, illustration, and convenience.
The following description is provided to assist the reader in gaining a comprehensive understanding of the methods, apparatuses, and/or systems described herein. Accordingly, various changes, modifications, and equivalents of the methods, apparatuses, and/or systems described herein will be suggested to those of ordinary skill in the art. Also, descriptions of well-known functions and constructions may be omitted for increased clarity and conciseness.
An apparatus and method described herein obtain information on a user's emotional state while a user is viewing content, build database using the obtained information on the user's emotional state, and search for and recommend emotion-based content based on the managed data.
Referring to
The emotion information acquiring unit 200 may extract information on a user's emotional state at the time of viewing content, and transmit the extracted information to the emotion information managing unit 300. In addition, the emotion information acquiring unit 200 may acquire information about a level of user satisfaction with recommended content after viewing the content, and transmit the acquired satisfaction information to the emotion information managing unit 300.
The emotion information managing unit 300 may manage the data input from the emotion information acquiring unit 200 on the individual content and user basis.
The emotion-based content recommending unit 400 may search for and recommend content based on the data managed by the emotion information managing unit 300 in response to a request from the user for content search and recommendation using the user emotion information as a keyword.
Referring to
The SNS data collecting unit 210 may receive information from each user, the information including user's SNS list information and ID information, and manage the received information. In addition, based on the information, the SNS data collecting unit 210 may search for and collect messages that a particular user has posted in all social network services for a predetermined period of time.
The emotion-descriptive word extracting unit 220 may extract words describing emotions from the data collected by the SNS data collecting unit 210.
The emotion-descriptive word classifying unit 230 may categorize emotional states into N groups including a sad group, a happy group, an angry group, a sensitive group, and the like, and create an emotion-descriptive word classification list including the categorized groups.
The emotion-descriptive word extracting unit 220 and the emotion-descriptive word classifying unit 230 may utilize a text mining research result.
The user satisfaction acquiring unit 240 may obtain information about user satisfaction with content that the user has watched in a specific emotional state.
Referring to
Referring back to
The creation of the classification distribution in 314 will be described in detail below.
Under the assumption that there are N emotional states and X emotion-descriptive words are extracted, the number of emotion-descriptive words included in an i-th emotional state is represented as xi. In this example, the classification distribution is expressed as a group of N data (α1, α2, . . . , and αN) each data representing a ratio of the emotion-descriptive words corresponding to each emotional state to the entire number of the emotion-descriptive words, and an i-th value αi is xi/X. In this case, X=Σi=1Nxi and 1=Σi=1Nαi.
In addition, after the completion of the classification distribution, the emotion information acquiring unit 200 transmits information about the user and the content watched by the user along with the calculated classification distributions to the emotion information managing unit 300.
The operations shown in
In another example, when the user requests the system for content search and recommendation based on an emotion-descriptive word, the system may store both request data and a recommendation result. In this case, the determination of whether the user has used the recommended content may be made by checking the stored data.
Referring back to
If the user does not provide the requested information, the operation ends. If the user provides the satisfaction level information, updating of the emotion information is performed while taking into account the satisfaction information.
Generally, the information received from the user is broadly classified into two groups, a satisfaction group and a dissatisfaction group, and satisfaction information of each group may be received.
Based on the received information, the user satisfaction acquiring unit sets a satisfaction weight (SW). If there are four levels of the satisfaction and dissatisfaction groups, there are four SWs w1, w2, −w1, and −w2. w1 and w2 represent levels, and sign “−” indicates dissatisfaction.
In addition, the user satisfaction acquiring unit 240 calculates the emotion classification distribution using the method used in operation 314. By multiplying the calculated emotion classification distribution by the set SW, a final classification distribution is calculated in 323. Thereafter, the user and content IDs and the final classification distribution value are transmitted to the emotion information managing unit 300 in 324.
The emotion information managing unit manages user IDs, content IDs, and emotion distribution values on a content-by-content basis and on a user-by-user basis. That is, the emotion information managing unit manages the information by conceptually dividing them into a content-based DB and a user-based DB.
The content-based DB manages information about overall emotion distribution history. The content-based DB comprehensively manages the total number of content uses and distributions of each of N emotional states on the basis of the content ID.
In contrast, the user-based DB comprehensively manages the number of uses of each content by each user and distributions of each of N emotional states on the basis of the user ID.
In this example, the total number of uses of content is increased by 1 in response to data incoming to the emotion information managing unit, resulting from the operations shown in
The emotion distribution history is managed by accumulating the distribution results from the operations shown in
Referring to
Generally, since the content-based DB contains combined data of a number of users having different characteristics, the included content may show distinct features and have even distribution values with respect to a particular emotion classification or emotion classification group.
Thus, content recommendation based on users having a similar emotional tendency is given a higher priority. To this end, the emotion-based content recommending unit finds users showing a similar emotional tendency to that of the input user in 530.
In order to find the similar users who are more relevant to the current emotional state of the user, the emotion classification distribution calculated in 520 is used rather than the overall similarity of the emotional tendencies. The methods for finding the similar users may be the same methods as used in social networking.
For example, the similar users may be found by using content information commonly used in the emotional states similar to the input emotion distribution. In response to the similar user information being found, the emotion-based content recommending unit mainly recommends content that the similar users have frequently viewed in the emotional state corresponding to the calculated emotion distribution in 540. Such content recommendation method may employ existing recommendation algorithms so as to improve recommendation performance.
However, the similar-user-based content recommendation requires a great amount of accumulated history information. Without a sufficient amount of collected data, there may be no, or at least few, recommendation results.
If only a few recommendations are made in 550 because of an insufficient amount of collected history information, a further recommendation of content having been frequently viewed and showing the emotion classification distribution that is similar to the distribution calculated in 520 is made based on the content-based DB in 560. Then, the user is notified of the recommendation result in 570.
An example of a method for calculating the distribution of the similar emotion classification may include a method of converting each emotion classification distribution and the distribution calculated in 520 into N-dimensional normalized vectors.
For example, if the input emotion classification distribution is (α1, α2, . . . , and αN) and an emotion classification distribution of content to be compared (or content viewed by a particular user) is (β1, β2, . . . , and βN), a normalized vector converted from each distribution may be represented as follows.
V
α=<α
V
α=<α
In this case, a degree of similarity between two emotion distributions may be calculated using the inner product of a vector or a distance between vectors depending on system performances. For example, in the case of use of a distance between vectors, since the similarity increases as the distance is shorter, the degree of similarity can be measured by the reciprocal of the distance between vectors. In contrast, in the case of use of an inner product of a vector, the inner product itself may be used as the degree of similarity.
According to the above exemplary embodiments of the present invention, a recommendation of multimedia content can be adaptively made based on a user's emotional state.
A number of examples have been described above. Nevertheless, it will be understood that various modifications may be made. For example, suitable results may be achieved if the described techniques are performed in a different order and/or if components in a described system, architecture, device, or circuit are combined in a different manner and/or replaced or supplemented by other components or their equivalents. Accordingly, other implementations are within the scope of the following claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 10-2012-0042255 | Apr 2012 | KR | national |
