KR20070086806A - Associative content retrieval - Google Patents

Abstract

A retrieval system is provided. First description data including dimension data for a first identified content item is extracted. This process may be repeated for additional available identified content items. Candidate description data is extracted. Then, a set of vector values for each candidate content item may be generated, each vector value representing a degree of similarity between the dimension data for a dimension, for example, metadata, usage history, genre, content type, of the first description data and the corresponding dimension data of the candidate description data. A similar candidate content item from the candidate content items may be selected based on the degrees of similarity represented by the generated set of vector values, and is provided.

Description

Associative content retrieval}

The present invention relates to the field of content retrieval, management and presentation.

The storage capacity of storage devices and databases, including hard drives on personal computers, and other types of storage media is growing rapidly in recent years. It is estimated that storage capacity is approximately doubled every 12 months, and network bandwidth is growing very quickly. As a result, storage devices store large amounts of content that are easily accessed by a user. Content that is not indexed or organized in a way that is transparent to the user may be "lost" and not retrieved until the user feels the need. On the other hand, a user may be overburdened by content stored on a storage device or database, and if the content is not managed or organized in some way to provide convenient access for the user, retrieve content available on a network such as the Internet. Can not.

There are various ways to retrieve data. Platt U.S. Patent Publication No. 2003/0221541 describes an automatic playlist generator in which multiple seed songs, including "undesirable seed" songs, are used to create songs on a playlist. Automatic playlist generator is disclosed in US Pat. No. 5,616,876 to Clluts for each song recorded in the epic by the editor, additional songs that are the first set of songs based on "style levels". However, none of these references disclose providing content to a user of a content type other than the content type of the user-designated identified content item.

Of course, it is also possible for a user to search for content items, but an attempt to find similar items can be a time consuming and cumbersome task, especially if the content type of the desired items is not known or specified to the user. In addition, retrieving content items becomes more difficult as content items are continuously accumulated in a storage device or database controlled by a user.

A method, system, device, apparatus, and computer readable medium are provided for implementing or executing the functions of a search system. First description data including dimension data for the first user-selected content item is extracted. Subsequently, candidate description data including corresponding dimension data for the candidate content item is extracted, and each candidate content item is of a content type different from that of the user-selected content item. A first set of vector values for each candidate content item can be generated, each vector value representing a degree of similarity between the dimensional data for the dimension of the first description data and the corresponding dimensional data of the candidate description data. Candidate content items from the candidate content items may be selected based on the degree of similarity represented by the first set of generated vector values. The selected candidate content item or item (s) is provided by a search system, through a user interface.

Dimensions A dimension of data is the content type of the item, the content style for the item, the genre of the item, the item metadata, the usage history of the item, the performer acting on the item, the director associated with the item, and the constructor associated with the item. Represents a rendering requirement for a creator or item. As used herein, the metadata may include a creation time of an item, a creation place of the item, an acquisition time of the item, and / or an acquisition place of the item.

The candidate content item may be selected only if the total degree of similarity represented by the first set of vector values exceeds a minimum threshold.

A candidate content item with the highest total degree of similarity may be selected as represented by the first set of vector values.

Additional content items can be identified. Description data is extracted that includes dimension data for a second identified content item grouped into a first identified content item. The candidate content item is then selected based on the second set of vector values that indicate degrees of similarity between the dimensional data for the second identified content item and the dimensional data of the similar candidate content item. Thus, the recommended content item may be selected such that the vector values of the first and second set of vector values are averaged, weighted averaged, or added.

A commonal vector can also be selected for weighted results. A common vector, the vector in which the dimensional data of the first identified content item is closest to the second identified content item, is selected, and when selecting a candidate content item, the value of the common vector is the first of the vector values. It can be weighted larger than the remaining vector values of the set and the second set of vector values.

Virtual content items may be configured. Description data is extracted that includes dimensional data for the first and second user-selected content items. Candidate description data including corresponding dimension data is extracted for the candidate content items, and each candidate content item is of a content type different from that of the user-selected content item. The virtual item may then be constructed by averaging or weighted averaging a virtual item set of vector values, wherein each vector value of the virtual item set of vector values is a dimension of the dimensional data of the first description data and a dimensional data of the second description data. The degree of similarity between the dimensions of. A set of vector values of each candidate content item can be generated, each vector representing a degree of similarity between the dimensional data for the dimension of the virtual content item and the corresponding dimensional data for the candidate content item. From the candidate content items, the candidate content item calculates one of the average, weighted average, and sum for each set of vector values of candidate content items as a testing value, and selects the candidate content item whose testing value exceeds a threshold. It may be selected by determining as the selected candidate content item. The selected candidate content item or item (s) is provided.

1 is a schematic illustration of both retrieval systems according to an embodiment of the invention.

2A-2C are flow diagrams illustrating operations of a system in accordance with the present invention.

3 illustrates a data chart of vector values according to an embodiment of the present invention.

The following discussion and the following drawings describe embodiments of the invention as best understood by the present inventors, but various embodiments of the invention are possible and the invention is capable of other ways without departing from the spirit of the invention It will be understood that it can be executed as. In addition, the features of the described embodiments can be optionally omitted and combined with other embodiments, or can be used to replace features or parts thereof in other embodiments without departing from the spirit of the invention. Therefore, the drawings and detailed description are to be regarded as illustrative in nature, and are not intended to limit the scope of the invention.

As shown in Fig. 1, the retrieval system 1-1 includes several modules described below. The modules or parts of the search system 1-1, and / or the search system as a whole may be comprised of hardware, software, firmware, or a combination thereof, while some modules may be comprised of hardware, for example, Other modules may be configured in software, firmware, or a combination thereof.

It should be understood that all of the modules of the search system need not be integrated or located with the same device. The deployed architecture is considered for the search system, which can "piggy-back" the appropriate modules provided by existing devices.

The following description relates to a retrieval system 1-1 which is physically integrated with the database 1-2 or connected via a wired or wireless connection. The database 1-2 may be implemented on a storage device such as a hard drive of a personal computer, a personal video recorder, an entertainment system, an electric organizer, a personal portable device, a jazz drive, or a disk drive. It can be implemented as a commercial storage server such as It will be appreciated that the database 1-2 may include multiple connected storage devices, capable of organizing or grouping content items on two or more such devices. In addition, the database is one such as disks, including CDs, DVDs, zip disks, floppy disks, data cartridges, etc., which can be loaded on and retrieved by the database 1-2. It can be understood to include the above storage media. However, it will be appreciated that the retrieval system 1-1 may also retrieve content via the network 1-9, such as LAN, WAN, Internet, and the like.

As shown in FIG. 1, the retrieval system 1-1 includes a description data extractor 1-11, which is a module that collects any types of data from a content item. The content item may be a video, video clip, movie, photo, text file, music data, audio file, or other type of multimedia data, JPEG file, or XML data. For example, the video may be a home video shot on a digital video recorder, and the movie may be commercially distributed film data, such as a film encoded as MPEG (including MPEG-2, MPEG-3, etc.). The photo may be digital photo data or a series of photos or a photo album, the text file may be a word processor produced file, a spreadsheet, or a computer code file, the music data may be an MP3, or the like. Can be.

The description data extracted by the description extractor 1-11 includes information about the content item. This description data describes the dimensions of the content item. These dimensions are:

Content types including media such as video, audio, photo, text files, and the like;

Content styles or genres, such as weekend movies, personal landscape photos, jazz music, and the like;

Metadata about the item, such as the creation time and / or location of the item, the acquisition time and / or location of the item;

Time and / or location of the playback and / or edit, final / first / middle, etc., the main time period of use (e.g., item is used primarily at night or Monday afternoon or 6-8 am, etc.), As a history of the use of the item, such as the time of acquisition of the item, the place of creation of the item, the place of acquisition of the item, the end use place, the main place of use (eg, the item is primarily used in the macro or mainly in the user's home, etc.); Such usage history data is sometimes known as metadata, and vice versa, usage history of the item, referred to as usage history data: and

It may include any at least one of an actor, director, creator, artist, actor, photographer, and the like associated with the content item.

This description data for the item may be the same as the World Wide Web (WWW), which is connected to the search system 1-1 from an item, from an index or database management file, or via a wireless or wired connection to the Internet 1-9. It can be found and extracted in a variety of ways, including from external sources.

The identified content item can be identified in various ways. The user can specify an item based on which other items, sometimes called "candidate content items", are searched for. Alternatively, the newly added or created content item may be automatically designated as the identified content item based on which other items are retrieved.

Based on these completed dimensions of the description data extracted by the description data extractor 1-11, the content item identifier 1-12 is the first identification via the network connection or with respect to the dimensions of these description data. The candidate content item is identified in the database from other sources similar to the content item. A vector constructor (1-13) generates a first set of vector values by assigning vector values to each of a plurality of vectors as follows: each vector corresponds to a dimension, and the value for the vector is a first value. The dimension of the identified content item reflects the degree of matching or similarity with the candidate content item.

For example, the vector corresponding to the style or dimension of the content item being genre has a high value if the identified content item and the candidate content item are the same genre, such as "Spanich holiday." A vector value of 1 or 0 may indicate little correlation or matching for a particular dimension between the first identified content item and the candidate content item, whereas a vector value of 9 or 10 may have a high degree of similarity or matching Can be represented. For example, when both content items have a genre of "Spanish Holidays", a vector (value 9 or 10) corresponding to the genre dimension is assigned. Alternatively, instead of using a scale of 1 to 10, the vector values may simply represent "strong", "normal", "weak" matching for the dimension. It will be appreciated that numerous other methods for these vector values may be used without departing from the spirit of the invention. The average or sum of the set of vector values for a pair of content items is calculated as the overall degree of similarity between the two content items.

If a second identified content item is available, the second set of vector values indicates that the second set indicates a degree of similarity between the second identified content item and the corresponding dimensions of the candidate content item. It can be similarly configured by the vector constructor 1-13 based on the description data extracted by the description data extractor 1-11 for the content item. In addition, there may be identified content items available. Therefore, this process of description data extraction and vector value set generation can be repeated for any number of identified identified content items 1-N, where N is a positive integer greater than one. Candidate content item selection is then performed based on all such generated vector value sets or their average.

If one or more identified content items are available, the common vector generation / threshold setter 1-14 may select one or more vectors whose vector values of the first set and the second set are consistently high. Such vector values may be weighted more than values for other vectors in the average or sum of a set of vector values representing the total degree of similarity between two items. In this way, a dimension that represents the identified first and second content items or tends to capture similarities between the identified first and second content items thus weights the group more than other vector values. It is characterized by being. Although shown as part of a single module 1-14, the individual modules, the common vector generator module and the threshold setter module may be configured as part of the search system 1-1, or these modules may be other modules. Can be included.

Virtual item constructors 1-15 are described under the background of discussion of the operation of embodiments of the present invention.

The content item selector 1-16 selects a candidate content item or items to be provided to the user. This module can also handle other tasks required for the operation of the search system, such as the overall control and characterization of the modules of the search system 1-1.

The search result output 1-17 interfaces with external devices and with other devices, including an interface with a user (not shown). In particular, the search result output 1-17 signals for the user interface of the content items retrieved by the search system 1-1. The user interface 1-3 may be a separate device or may be integrated with another device or system, such as a personal computer or a personal video recorder, or one or more storage devices or other devices described above.

Operation of an embodiment of the present invention is described with reference to FIGS. The first content item is identified by the user via the user interface 1-3 shown in FIG. 1 as described above, or is isolated content in a newly added content item or database 1-2, for example by the system. Automatically identified by the detection of the item.

The description data extractor 1-11 of the retrieval system 1-1 extracts the first description data for the first identified content item, as mentioned in S2 of FIG. 2. 3 shows boxes 6-11 labeled with reference to the identified content item 1. In S2, the dimension data for each of the dimensions for the identified content item is compiled. Depending on the needs of the user, it will be appreciated that some or all of the dimensions identified above may be more appropriate, while others may be completely inappropriate or not used by the search system according to the present invention. In addition, other dimensions not explicitly mentioned herein may be particularly suitable and may be used by the retrieval system 1-1.

If the identified additional second content item, shown in FIG. 3 as 6-12, is available or identified, steps S3 and S4 are performed: At S3, the description data for the identified content item is extracted, and at S4 The dimension data for each of the dimensions for the second identified content item is edited. As shown in FIG. 3, multiple content items may be identified as the basis of content search. 3 shows a first identified content item 6-11, a second identified content item 6-12, and an identified content item N, 6-14. Therefore, this process is repeated for each of the first N identified content items.

The content item identifiers 1-12 of FIG. 1 identify candidate content items in the database 1-2 via the network or elsewhere, while the description data extractor 1-11 (FIG. 2A), at S5, The description data for each of the candidate content items is extracted, and in S6, the dimensional data for each of the content items is edited. The process of extracting the corresponding description data of the second candidate content items is performed at S7, and the editing of the dimensional data for the second candidate content item is performed at S8.

According to an aspect of the present invention, in S9, depending on the system settings or depending on the user's setting or the current command, it can be determined that the virtual item is configured as a basis for determining similarity of candidate content items, The processing will proceed as shown in 2c. Otherwise, processing proceeds as shown in FIG. 2B.

Based on matching or similarity between each dimension of each identified content item and the corresponding dimension of each candidate content item, the vector value is constructed by the vector constructor 1-13, as shown in S11 of FIG. 2B. . 3 shows a set of vectors 6-3 with values reflecting the degree of similarity with respect to the corresponding dimensions of the first candidate content item 6-21 and the first identified content item 6-11. The table 6-1 is shown together. Similarly, the set of vector values 6-4, together with the second candidate content item 6-22, reflects the similarity of the dimensions of the first identified content item 6-11. For the second identified content item 6-12, the set of vector values 6-5 reflects the degree of similarity for the corresponding dimensions with the first candidate content item 6-21, and On the other hand, the set of vector values 6-6_ reflects the degree of similarity between the dimensions of the second identified content item 6-12, along with the candidate content item 6-22.

Each set of vector values may also include an average vector value determined at S12, based on the algebraic average, mode, median, or summation calculation of the vector values of this set, reflecting the average similarity for the pair of content items. have. Thus, for example, the vector values 6-3 of FIG. 3 may include a first vector value, a second vector value, and an average value for the h-th vector value and the set.

Further identified content items are also available, and the process continues to extract the dimensional data and find a set of vector values based on the similarity of the candidate content items with the corresponding dimensions. Boxes 1-14 of FIG. 3 refer to the identified content item M.

According to an embodiment of the present invention, in S13, the common vector value set is determined based on the similarity of the dimensions between the identified content items. Therefore, the most similar dimensions are identified, and representative vectors can be weighted or used exclusively than other vectors. In this way, the dimension that represents the identified first and second (additional) content items and tends to capture similarities between the identified content items is thus weighted such that the group formed is more weighted than other vector values. , Exclusively used to determine similar candidate content items.

In S14, by averaging or adding the corresponding vector values of the candidate content item 6-21, an additional set of vector values 6-8 reflecting the overall similarity for each of the dimensions for each candidate content item is calculated. Can be. Therefore, the overall similarity with the identified content items for the dimension, for example, by adding or averaging the corresponding vector values for each set of vector values for the candidate content item (for column 6-2). The degree is achieved for the first candidate content item. In addition, all of the vector values of set 6-8 may be added or averaged to obtain the total similarity value for the candidate content item.

It will be appreciated that the mean used herein may include an algebraic mean, mode, intermediate or other statistical function appropriately selected to provide a composite view of the selected values. In addition, this statistical function as well as the sample sum of the values can be used. Depending on the type of content item and depending on the needs of the database and the user, any dimensions of the content item may be more important than anything, and for this reason it is advantageous to weight the vectors corresponding to any dimensions above all. Can be. The degree to which these factors are weighted depends on the needs of the application and the user.

Once the vector values of the entire similarity set 6-8 are generated, the minimum similarity threshold can be used to remove dissimilar candidate content items, as shown in S15 of FIG. 2B.

It is also contemplated that different thresholds may be used for various vectors, depending on the user's needs and application. Thus, candidate content items whose vector values meet or exceed a threshold value are grouped into content items identified by group organizer 1-17, while other candidate content items are rejected. Alternatively, the most similar candidate content item or a predetermined number of most similar candidate items may be selected for grouping into identified content items, while the remaining candidate content items may be rejected.

According to an aspect of the present invention, the retrieved content item is of a different content type than the content type of the user-selected content item. For example, the user-selected content item may consist of a type music file or an MP3, and the retrieved content item may consist of content type photo data. In this way, for example, pictures of any genre can be searched to match user selected music of the same genre.

These selected candidate content item (s) are provided to the user or to the user interface 1-3 at S16. Signals may be provided directly to the database 1-2 to cause retrieval of selected candidate items into the database or user interface 1-3. A notification is provided to the user interface 1-3 to notify the user (not shown) of the searchable content item. The notification may consist of an identification of the content item to be retrieved, a description of the content item, a URL or link to the content item, a search of the entire content item or part thereof, or a combination thereof. In S17, the processing ends.

2C illustrates further processing in accordance with an aspect of the present invention using a virtual content item. In S21, the virtual item constructor 1-15 analyzes the dimensions of the identified content items based on which grouping is sought. In S22, a representative content item for all identified content items called virtual content items 6-15 is constructed based on the average or weighted average dimensions of the identified content items. For example, if all of the identified content items consist of the genre "Spanish Holiday," then the virtual content item also has the genre "Spanish Holiday." Subsequently, in S23, sets of vector values 6-7 are generated based on the similarity of the candidate content items with the dimensions of this virtual content item. In S24, the threshold is applied at the time of selection, similar to when candidate content items are selected or the candidate content item or items with the highest score are selected.

Based on the similarly selected candidate content items using the threshold, or based on the most similar predetermined number of candidate content items selected, in S25, as discussed, a notification signal is output from the search results (1-17). Is provided by In S26, the process ends.

For example, a user may edit digital data representing photos of recent holidays in Spain in a database, and search for other content items with Spanish themes available in the database on another connected storage medium or via the Internet available. Suppose that The user can select three photos as each of the identified content item 1, the identified content item 2, and the identified content item 3 via the user interface 1-3. The search system searches for a data file representing Spanish music found as the selected candidate content item. The user may not remember the existence of Spanish music and may not remember where to find it in the database 1-2, and in fact the data file is accessed by another user who has access to the database 1-2. Can be added or retrieved by the retrieval system 1-1 from another storage device or from the world wide web. In any case, the user is notified of the retrieved content item, and / or the retrieved content item is associated with the user selected content items. The user may be accompanied by viewing of Spanish holiday photos along with Spanish music.

Embodiments of the invention provided in the foregoing description are merely examples. However, the scope of the present invention is provided in the claims.

Claims (17)

In the content search method, Extracting first description data including dimension data for a first user-selected content item (S1); Extracting candidate description data including corresponding dimension data for candidate content items (S5), wherein each candidate content item is a content type different from the content type of the user-selected content item; Extracting candidate description data (S5); Generating a first set of vector values for each candidate content item, wherein each vector value is the dimension data for the dimension of the first description data and the corresponding dimension data of the candidate description data. Generating (S11), indicating a degree of similarity between; Selecting a candidate content item from the candidate content items based on the degrees of similarity represented by the generated first set of vector values (S15); And Presenting the selected candidate content item (S16). The method of claim 1, wherein the dimension of the dimension data is the content type of the item, the content style of the item, the genre of the item, the usage history of the item, the performer acting on the item, the director associated with the item. (director), a creator associated with the item, rendering requirements for the item, and any metadata for the item. The method of claim 2, wherein the metadata includes a creation time of the item, a last use time, a main use time period, an acquisition time of the item, a place of creation of the item, a place of acquisition of the item, a place of end use, and a main use place. A method of content search, indicating one of the. 2. The method of claim 1, wherein the candidate content item is selected only if the total degree of similarity is above a minimum threshold as represented by the first set of vector values. 2. The method of claim 1, wherein a candidate content item with the highest total similarity degree represented by the first set of vector values is selected. 2. The method of claim 1, wherein extracting description data comprising dimension data for the Nth identified content item grouped into the first identified content item (S3), wherein N is an amount greater than one. Extracting the description data, which is an integer (S3); And Automatically selecting (S15) the candidate content item based on an Nth set of vector values indicating degrees of similarity between the dimensional data for the Nth identified content item and the dimensional data of the similar candidate content item; And a content retrieval method. 7. The method of claim 6, wherein the candidate content item is selected such that the first set of vector values and the Nth set of vector values are one of average, weighted average, and added. 7. The method of claim 6, further comprising: selecting as a common vector a vector whose dimension data of the first identified content item is closest to the Nth identified content item, wherein the first set of vector values; Selecting the candidate content item by weighting the common vector value above the remaining vector values of the Nth set of vector values. In the content search method, Extracting first description data including dimension data for the first user-selected content item (S1); Extracting Nth description data including dimension data for an Nth user-selected content item (S3), wherein N is a positive integer greater than one (S3); Extracting candidate description data including corresponding dimension data for the candidate content item (S5), wherein each candidate content item is extracted from the candidate description data, the content type being different from the content type of the user-selected content item; Step S5; Computing a virtual item by one of averaging and weighted averaging a virtual item set of vector values, wherein each vector value of the virtual item set of vector values is a value of the dimensional data of the first description data. Calculating (S22) the virtual item, indicating a degree of similarity between a dimension and a corresponding dimension of the dimensional data of the Nth description data; Generating a set of vector values for each candidate content item (S23), wherein each vector value is determined by the similarity between the dimensional data for the dimension of the virtual content item and the corresponding dimensional data for the candidate content item. Generating (S23) the set of vector values, the degree of which; A candidate content item is selected from the candidate content items by calculating one of an average, a weighted average, and a sum for each set of vector values of the candidate content items as a testing value, wherein the testing value exceeds a threshold. Determining the candidate content item as the selected candidate content item (S24): and Presenting the selected candidate content item (S25). In a content retrieval system, A description data extractor (1-1 configured to extract first description data comprising dimensional data for a first user-selected content item and also configured to extract candidate description data comprising corresponding dimensional data for candidate content items 11) the description data extractor (1-11), wherein each candidate content item is a content type different from the content type of the user-selected content item; A vector generator (1-13) configured to generate a first set of vector values for each candidate content item, each vector value being the dimension data for the dimension of the first description data and the candidate description data; Said vector generator (1-13), indicating a degree of similarity between corresponding dimensional data; A content item selector (1-16) configured to select a candidate content item from the candidate content items based on the degree of similarity represented by the generated first set of vector values; And A search result output (1-17) configured to provide the selected candidate content item. 11. The method of claim 10, wherein the dimension of the dimension data is the content type of the item, the content style of the item, the genre of the item, the actor acting on the item, the director associated with the item, the creator associated with the item, the item And one of the rendering requirements for the item and any metadata for the item. 12. The method of claim 11, wherein the metadata is the creation time of the item, the last use time, the main use time period, the acquisition time of the item, the place of creation of the item, the place of acquisition of the item, the last place of use, the main place of use A content retrieval system. 11. The content retrieval system according to claim 10, wherein the content item selector (1-16) is configured to select the candidate content item only if the total degree of similarity represented by the first set of vector values exceeds a minimum threshold. . 11. The content retrieval system according to claim 10, wherein the content item selector (1-16) is configured to select the candidate content item having the highest total similarity degree as represented by the first set of vector values. 11. The description data extractor (1-11) according to claim 10, wherein the description data extractor (1-11) includes dimensional data for an Nth identified content item (N is a positive integer greater than 1) grouped into the first identified content item. Configured to extract the description data, The content item selector 1-16 further selects the candidate content based on the Nth set of vector values indicating a degree of similarity between the dimensional data for the Nth identified content item and the dimensional data of the similar candidate content item. And select an item automatically. 16. The apparatus of claim 15, wherein the content item selector (1-16) is configured to select the candidate content item such that the first set of vector values and the Nth set of vector values are averaged, weighted average, and added. A content retrieval system. 16. The common vector generator / threshold setter as claimed in claim 15, wherein the dimensional data of the first identified content item is configured to select as a common vector a vector representing a dimension closest to the Nth identified content item. 14), The content item selector 1-16 is configured to select the candidate content item based on weighting the value of the common vector above the remaining vector values of the first set of vector values and the Nth set of vector values. , Content retrieval system.

Images