WO2006090314A2

WO2006090314A2 - Online matching system

Info

Publication number: WO2006090314A2
Application number: PCT/IB2006/050521
Authority: WO
Inventors: Mauro Barbieri; Nevenka Dimitrova; Lalitha Agnihotri
Original assignee: Koninklijke Philips Electronics N.V.
Priority date: 2005-02-22
Filing date: 2006-02-17
Publication date: 2006-08-31
Also published as: WO2006090314A8

Abstract

The present invention relates to a system and method for matching user profiles owned by persons, or for matching user profiles owned by persons with user profiles owned by groups or organizations. In particular the present invention provides a solution that matches user profiles more accurate since the system extracts data that can not be tampered by the owner of a user profile. Thus the user profile comprises both information provided by the user and information provided automatically by the system.

Description

Online matching system

The present invention relates to an online dating system, in particular it relates to an online matching system for automatically providing matches between user profiles.

In the last few years Internet related technologies and other communication technologies have made it possible for people to meet and communicate with other people over the Internet. Because of this recent trend a multitude of online communities and dating services has developed for making it even easier for people to interact with a vast amount of other people. However, due to the huge amount of information accessible on the Internet, it may be very hard for people to find other people that would be suitable to interact with. Furthermore, the increase in work pace/load makes it harder for people to combine their private life with their professional life, thus less time can be set a side for finding a potential partner. Since it is important for human beings to engage in relationships, there is a need for technical solutions that can help people to combine their private life and professional life for increasing the chances of meeting a potential partner and maybe engage in a relationship.

On-line dating services help people to find new partners that match their interests with the aim of building a relationship.

However, the vast majority of current online dating or matching systems rely on detailed users' profiles created using various types of questionnaires. Users need to fill in a questionnaire and the results are used to create profiles upon which possible partners are matched. One disadvantage of existing online dating systems is that they are based on information explicitly provided by users and therefore not guaranteed to be correct. Users can consciously or unconsciously cheat or provide erroneous information and their profile will not correspond to reality. Furthermore, a user of these systems needs to spend time creating a user profile, time which may be a sparse asset to some people.

One service for matching persons based on TV programs is the TV-friendship system. However in this service the user needs to manually mark which TV programs that he/she likes. Thus none of the above problems are solved regarding input of erroneous information and/or save time for a user.

Thus it is preferred to provide a solution for improved matching of profiles in an online dating service. It is an advantage achieved by the present invention to provide a solution that increases the ease of use for a user.

It is an advantage achieved by the present invention to provide a solution that saves time for a user.

It is further an advantage achieved by the present invention to provide a solution that overcomes faulty inputs.

It is further an advantage achieved by the present invention to match people from different cultures.

It is a further advantage achieved by the present invention to match users such as potential candidates to a job position, group, or organization. According to a first aspect of the invention the above object and advantages are achieved by a method of providing matches on an online dating system, said system comprising: a first user profile, and at least one second user profile, the profiles each further comprising: a personal profile comprising at least one personal property comprising a value, a partner profile comprising at least one partner property comprising a value and - a viewing profile, the method comprising steps of: comparing the partner property values in the partner profile in the first user profile with the personal property values in the personal profile in the second user profile, comparing the partner property values in the partner profile in the second user profile with the personal property value in the personal profile in the first user profile, establishing a match when, upon comparison, at least a pre-determined number of the properties of the first and second user profile have substantially same values, - comparing the viewing profile in the first user profile with the viewing profile in the second user profile in order to calculate a match rate, and when the match rate exceeds a predetermined threshold, providing at least one of the first user profile and second user profile with a matching message.

Thus the above method has the advantage that an online dating system/service is able to provide a more accurate match between user profiles. Since the method preferably also uses the viewing profile in the match it is possible to avoid or decrease the risk of matches based on faulty or erroneous inputs. Furthermore it is possible to match user profiles without inputs from a user.

In a second aspect of the invention, the above and other objects are fulfilled by a method of creating a mapping table in an online dating system comprising: a user profile, at least one viewing history, and at least one recorded data set, the at least one viewing history comprising (EPG) high-level features related to the recorded data sets, the user profile further comprising: at least a viewing profile, the method comprising steps of: extracting low- level features from the recorded data sets, - grouping the data sets, based on the low-level features, assigning each group one or more label(s) associated with the low-level feature(s), based on the low-level features and the high-level features, deriving combined labels related to the group so that the group is defined by the combined labels, - matching the viewing profile in a user profile with the combined labels in order to map suitable data sets, and storing these matches in a mapping table. The above method has the advantage that an online dating service is able to derive a mapping between the viewing profile in a user profile and personal genres defined by the combined labels.

Preferably the personal genres are the groups of data sets defined by the combined labels.

Furthermore the above method has the advantage that an online dating system/service is able to generate personal genres that are much more descriptive than just the viewing history, since the labels comprises both high-level features and low-level features. The recorded data set preferably relates to content data that a user wants to enjoy. The recorded data set may be stored on an online database and being associated to the user, that has selected the content data.

The recorded data set may also be stored on an electronic device located at the location of the user. The viewing history preferably comprises Electronic Program Guide (EPG) high-level features associated with the recorded data sets, information such as title, genre, cast, director, time, date and so forth.

The viewing history may be updated with new information when a user selects a data set that the user wants to watch. The selected data set is referred to as the recorded data set.

Clustering operations may perform the grouping of data sets.

In a third aspect of the invention, the above and other objects are fulfilled by a data structure comprising: a user profile, the user profile comprising: a personal profile, a partner profile and a viewing profile.

The above data structure has the advantage that an online dating system/service may be able to transfer detailed information in a structured way so as to facilitate communication and internal processes that finally result in accurate matching of user profiles.

In a fourth aspect of the invention, the above and other objects are fulfilled by a computer system comprising: a server for storing at least one or more user profiles in a database, the server being connected to a network, an electronic device for downloading data sets and for downloading at least parts of at least one of the one or more user profiles from the database via the network, each user profile comprising: a personal profile, a partner profile, and a viewing profile, said computer system being programmed to: - compare a first user profile with one or more second user profiles based on at least one of the personal profile and the partner profile and the viewing profile, upon match, sending a matching message to at least one of the first user profile and second user profiles.

The above computer system has the advantage that an online dating system/service is able to provide a more accurate match between user profiles. Since the system preferably also uses the viewing profile in the match it is possible to avoid or decrease the risk of matches based on faulty or erroneous inputs. Furthermore it is possible to match user profiles without inputs from a user.

Thus the present invention allows for matching data, preferably data relating to user profiles, more accurate in an on-line dating system or on any kind of electronic device such as servers, set top boxes, mobile phones, PDAs, DVD player, game console, etc., the electronic device being able to execute necessary steps according to the present invention.

Furthermore the present invention is able to create and send matching messages to relevant electronic devices, associated with a matched user profile, or to another location/electronic device where the holder of a user profile is able to receive and view the matching message, e.g. an online web account.

By the present invention user profiles are preferably being matched according to some predetermined thresholds, rules, or rules derived by data mining by using different techniques such as regression, classification and clustering/grouping or by other statistical calculations.

In a further embodiment the viewing profile, in the user profile, may further comprise: a like list and a dislike list, the like list and dislike list preferably comprising data sets , the method further comprises steps of: comparing data sets in the like list in the first user profile with data sets in the like list in the second user profile to calculate a like rate, - comparing data sets in the dislike list in the first user profile with data sets in the dislike list in the second user profile in order to calculate a dislike rate, comparing at least one of the like rate and/or dislike rate to at least one predetermined threshold, and if at least one of the like rate and/or dislike rate exceeds the predetermined thresholds, sending matching messages to the first user profile and second user profile.

Preferably the like list and dislike list comprises list elements comprising the data sets and/or metadata related to the data sets and/or pointers pointing at the data sets.

The data sets may comprise content data, where content data may be any kind of data that can be displayed/played to a user such as sounds, movies, text, pictures, 3D world, virtual reality environments, TV programs, music or other viewable/audible interactive content. Preferably the content data is tagged with metadata.

The like list and dislike list preferably comprises list elements. However, the like and dislike list, may comprise any other suitable data structure such as an array, queue, tree, table and so forth. Which data structure to be used may depend on the application and context wherein the present invention will be used.

Hence, the word "list" in the present document does not exclude that the underlying data structure in the like and dislike list may be another data structure such as an array, queue, tree, table and so forth. In a further embodiment the data sets may comprise high-level features and low-level features, the method further comprises steps of: extracting high-level features from the data sets in at least one of the like list and/or dislike list, extracting low- level features from the data sets in at least one of the like list and/or dislike list, creating a like group comprising high-level features and low-level features from the like list, creating a dislike group comprising high-level features and low-level features from the dislike list, comparing the like and dislike groups of the first user profile with like and dislike groups in one or more second user profiles in order to calculate at least one feature rate, and comparing the feature rate with at least one predetermined feature threshold, and upon approval, sending matching messages to the first user profile and second user profile.

Preferably the high-level features are chosen from a group of features comprising: - title, genre, cast, director, time, date, broadcaster, language, country, year of production, certification, such as parental certification, ratings and so forth.

The like and dislike groups may be incorporated into the user profile in order to obtain an even more detailed description of the user profile.

Preferably approval means that the comparison of the related features is under (or above) a certain threshold (similar values of the features). The threshold may be a lowest or highest limit depending on the context. For example if the threshold relates to a like list it may be the lowest limit, and if it relates to a dislike list it may be the highest limit. Furthermore threshold windows may be created by a lowest and highest limit.

Preferably the low- level features are chosen from a group of features comprising: audio features, such as: audio loudness; percentages of speech, music, noise, silence, applause, laughter, and other categories in which the audio signal can be automatically classified. video features, such as: film pace (combination of shot cut rate, motion activity and audio loudness). textual features, such as: language type (feature extracted from the speech transcript or closed captions or subtitles) as formal, informal, etc. Furthermore textual topics, such as: politics, comedy, health, psychology, etc. may also be extracted. scene features, such as: percentage of indoor/outdoor scenes (indoor/outdoor classification can be done automatically based on color/luminance and edge features); presence of overlaid text and graphics. Outdoor/indoor features: where the shots are automatically labeled as either indoor (home/office or public/private environment) or outdoor e.g. city/landscape. mood tone features, such as: anger, fear, joy, acceptance, rejection, surprise, expectancy, etc. human features, such as: percentage of scenes containing human faces (computed using face detection) percentage of scenes containing humans (computed using face detection and skin detection); number and gender of speakers (using automatic speaker change detection and speaker gender classification), and camera motion features, such as: amount and statistics about camera motion, panning, zooming, tilting, etc.

The above high-level and low-level groups are not exhaustive, the groups may further comprise other features not mentioned herein such as features related to games, web pages and other online content.

Preferably the high-level and low- level features comprise attribute-values that have been automatically derived from a data set or the alike by a software module located in an electronic device being able to execute a computerized method for performing different signal/audio/video detection, classification and analysis techniques.

Preferably the high-level features and low- level features are computed from the audiovisual signal (including attached textual information such as closed captions etc.) by content analysis algorithms (e.g. face detection, indoor/outdoor classification, etc.).

In a further embodiment the online dating system may comprise a database comprising couple histories, the couple histories preferably comprising matched user profiles where the users are still together, the method comprises steps of: from the matched user profiles, extracting matching patterns relating the matched user profiles, storing the extracted patterns in the database, - creating matching rules from the stored patterns, and using the matching rules for comparing a first user profile with one or more second user profiles.

Thus by storing these patterns it is possible to perform data mining and extract preferred matches. Thus the match will be based on a more reliable base and provide users with a better service.

However first the system may go through a training period in order to create the database comprising couple histories. The system will thereafter be able to extract patterns and create matching rules, and finally use these for matching. In a further embodiment the user profiles may comprise at least one recorded data set, the method comprising steps of: finding second user profiles comprising the same recorded data set as the first user profile, - sending a viewing proposal to the first user profile and at least one second user profile having the same recorded program, so that the profiles can view the recorded program at the same time.

In a further embodiment, at least one of the personal profile and partner profile in the data structure comprises at least one of the following properties: name, gender, age, nationality, occupation, income range, ethnic background, sports, arts, area of expertise, experience related to job, knowledge, profession, personal attributes, and personality traits and so forth.

Wherein personal attributes may comprise properties such as: hair color, eye color, iashionable, athletic, attractive, caring, creative, determined, motivated, honest, generous, mysterious religious, intelligent, artistic and so forth.. Personality traits may comprise properties such as: mood, self-esteem, introvert/extravert, sensation/intuition, self- monitoring and so forth.

Furthermore the viewing profile in the data structure may further comprise at least one of a like list and a dislike list. At least one of the like list and dislike lists may further comprise at least one of high-level features and low- level features.

In another embodiment the computer system according to the invention may further be programmed to match user profiles according to the first aspect of the invention.

The invention may be implemented by software on a data carrier executable on computing hardware for executing the method according to any of the methods described above.

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

Fig. 1 is a schematic illustration of a first and second user profiles.

Fig. 2 is a schematic illustration of a viewer profile comprising a like list and a dislike list.

Fig. 3 illustrates a flowchart for association of personal characteristics and personal genres. Fig. 4 is a schematic illustration of a computer system according to the present invention.

Figures are preferably schematically drafted in order to facilitate the understanding of the invention. Therefore other designs that could be drafted in the same schematic way are implicitly also disclosed in this document.

In this application some specific terms are used, below follows a description of these. The following section is preferably read with reference to the figures 1,2 and 3. User Profiles - UP

The first and second user profile preferably relates to a user/person. The user profiles may comprise information about the user (Personal Profile) and his/her desire relating to a potential partner (Partner Profile). Thus a person is able to fill in some information about him/her self and also information about a potential partner such as personal characteristics, look and so forth.

The user profiles further comprises a section, preferably not accessible by the user. This section is the viewing profile comprising data extracted from a data set such as a movie, game or any other content data.

In a second embodiment the first user profile may relate to a user such as a person and the second user profile may relate to an organization such as a company, or a department in a company or a job position and so forth. Thus the second user profile may comprise data relating to persons in an organization. In this embodiment the invention may be used to match a potential employee with a potential employer, or to match groups within an organization, experts, contact persons and so forth. Thus characteristics of a group may also be extracted by the use of the present invention. Of course, in this case certain type of information might be considered private in some cultures and therefore not required to be present in the profiles.

Hence by using the present invention an organization is able to find characteristics of different groups/employees and is thereby able to find for example a suitable contact person between two groups within a larger organization or between different cultures if it is a global company, or a new employee.

Personal Profile - PeP

Relates to the user (holder) of the user profile and thus preferably describes the user. For example the personal profile may contain a list of personal properties relating to the user as described earlier in this document. The properties in the personal profile preferably comprise a value.

Partner Profile - PaP

Relates to information about a potential partner, and thus preferably comprises a list of partner properties relating to preferred characteristics of a potential partner such as personality and look according to the user (holder) of a user profile. The properties in the partner profile preferably comprise a value.

Viewing History - VH

Preferably comprising high-level features from electronic program guides associated with content data such as movies, games and so forth.

Couple Histories

Preferably comprises successfully matched user profiles where the holders/users of the profiles are married or at least still together and actively dating. From the couple histories, patterns can be extracted and analyzed for being used when matching other user profiles.

Viewing Profile - VP

Is a part of a user profile and may comprise information about content data such as movies. The viewing profile preferably contains information of which movies the holder of the user profile has watched. Furthermore the movies in the viewing profile may be divided into a like list (L) and a dislike list (DL).

List f Like List (D and Dislike DL)

Preferably the viewing profile further comprises a like list and dislike list. After a user has enjoyed content data the user may be asked if he/she liked it or not. This may be achieved by providing a selection window on the screen/display where the content data has been shown. The selection window gives the user the option of selecting if he/she liked it or not. If the user liked it, the content data is added to the like list and if the user did not like it, the content data is added to the dislike list.

Furthermore the data sets in the like list and dislike list preferably may comprise high-level and low- level features as will be described below. High-level features

As described above the high-level features preferably relates to title, genre, cast, director, time, date, broadcaster, language, country, year of production, certification, such as parental certification, ratings and so forth. The high-level feature is preferably a part of a data set, such as metadata describing the data set. This metadata may be extracted automatically and stored in a user profile in the like list (L) or dislike list (DL) as will be explained later.

For example the movie "Fight Club" has the following high-level features: Title Fight Club

Genre Drama/Thriller/Mystery/Crime

Cast Edward Norton, Brad Pitt etc.

Director David Fincher

Time 139 Min Producer Art Linson, Ross Bell

Date 12/05/2015

Broadcaster 20th Century Fox

Language English

Country USA Year of production 1999

Certification Argentina: 18,..., Sweden 15,...

Ratings 8,5/10

The high-level features may be stored in strings and thus extracted by reading the strings. Low-level features

As described above in this document the Low-level features may be classified into the following classes: audio features, video features, textual features, scene features, mood tone features, human features, and camera motion features.

The low- level features demands more processing time and processing power in order to be extracted from a data set.

Audio features

For example the audio features may be extracted by analyzing the audio signal in order to measure the loudness, percentage of speech, music, noise, silence, applause, laughter and so forth. For example, a method to extract audio features such as silence from a compressed audio signal comprising blocks of quantized samples wherein a given block is provided with a set of scale factors, may be performed by extracting the set of scale factors from the compressed audio signal, and estimating the signal power in the given block based on a combination of the scale factors. This and other methods are well known to persons skilled in the art and thus not further explained.

Furthermore the following documents discloses features and methods to classify audio into classes, such as speech, music, noise, silence and so forth. McKinney, M.F. and Breebaart, DJ. , "Features for audio and music classification", proceedings of ISMIR 2003, 4th Int. Conf. Music Inform. Retrieval, Washington DC, USA, hereby incorporated by reference. Video features

The Video features may be extracted by measuring and analyzing the film pace, which may be a combination of measuring shot cut rate, motion activity and audio loudness.

Example of how cut-rate may be measured and extracted Cut-rate measures the frequency of shot-cuts in a video. A shot is a contiguous camera take. For a certain program (or part of it) "cut-rate" can be defined as ratio between number of shot cuts and program (or part of it) duration. Alternatively it can also be defined as inverse of average shot duration. The idea behind it is that shot duration (and thus cut-rate) influences our perception of "pace" (how "fast" or "slow") of a video. Short shots (high cut- rate) are typical of last paced videos such as action films, or music video clips while long shots (low cut-rate) are used in slow paced videos such as drama or documentaries. Motion activity extraction may be performed by motion vector based measures of motion activity. For example a number of low complexity measures computed from compressed domain MPEG block motion vectors may be measured, such as to measure the average of motion vector magnitudes or the median of the magnitudes.

Textual features The textual features such as language type may be extracted by using speech recognition, analyzing speech transcript or closed captions or subtitles, as formal, informal and so forth.

Textual information obtained from the closed caption, speech transcript or speech recognition system is classified in predetermined categories (e.g. formal, informal, etc.) based on the number of co-occurrences of predetermined keywords. Alternatively a statistical classifier (e.g. support vector machine) can be trained by example to classify text as belonging to one of the predetermined categories. Scene features The scene features may be extracted by analyzing content data such as performing image analysis on the frames in a movie in order to extract the percentage of indoor/outdoor scenes. Classification can be done automatically based on color/luminance and edge features; presence of overlaid text and graphics. Outdoor/indoor Scene features may furthermore be extracted where the shots are automatically labeled as either indoor (home/office or public/private environment) or outdoor e.g. city/landscape.

Text overlay detection may be performed by first detecting the overlaid text, this may be performed by detecting rectangular bounded areas of arbitrary size and position within the video frames. Thereafter the overlaid text may be extracted by transformation of the detected areas with text into binary images where all pixels that do not belong to characters are discarded. The binary image may then be converted into text of a computer readable format and finally recognized.

Another technique for detecting text in video segments comprises seven steps: Channel Separation, Image Enhancement, Edge Detection, Edge Filtering, Character Detection, Text Box Detection, and Text Line Detection.

Indoor/outdoor classification:

An approach to Indoor/Outdoor scene classification may be to analyze video frames and extract color and texture features and represent them by vectors. Thereafter a Support Vector Machine (SVM) may be trained on images in a database in order to be able to provide classification values based on the color and texture feature vectors. Based on the values a second SVM may be used to produce the final indoor/outdoor classification result. This and other methods are well known to persons skilled in the art and thus not further explained. Furthermore other techniques and methods may also be used in combination with the present invention.

Mood tone features

Mood tone detection of the movie/program, such as anger, fear, joy, acceptance, rejection, surprise, expectancy, etc.

Extraction of mood tone may for example be performed by extracting color from video frames. Most colors are associated with a mood, for example black may be associated with hatred, mourning, sorrow; Red may be associated with Love, Life, Noble and so forth. Therefore video frames may be analyzed by known methods and image analyses techniques in order to extract dominant colors.

Human features The human features may be extracted by using face detection, skin detection, voice detection and speaker gender classification.

Face detection and person identification:

Face and person detection may be performed by known image analyzing methods and techniques. Furthermore face detection may be combined with speaker identification, thus also audio data may be analyzed in order to identify a person.

Face detection may be performed by detecting skin regions over an entire image, and then generates face candidates based on the spatial arrangement of the skin patches. The algorithm may constructs eye, mouth, and boundary maps for verifying each face candidate. These techniques are widely used and known to the person skilled in the art. Camera Motion features

The camera motion features may be extracted by analyzing each frame and compare it with the previous and/or subsequent frames in order to calculate the camera motion. In this way camera motion such as panning, tilting, rolling and zooming can be extracted.

Since the low- level features demands quite much processing power, the Low- level features are preferably extracted once from a data set, and stored in a database for later retrieval.

However the extraction of low- level features may also be performed online in real-time, for example during streaming of a movie from a web site or while watching the movie.

Additionally low- level features for certain content could also be obtained by a service provider for example in MPEG-7 format or the alike.

Like group (LG) and Dislike group (DG) May be stored in the like list or dislike list in the viewing profile. Preferably the like group and dislike group comprises high-level features and low-level features associated to a specific data set.

For example in the case high-level and low-level features are extracted from a movie that the user has defined as a movie he/she liked, the features are preferably stored in the like group. Thus, the movie fight club may have the following like group:

Luce Group

High-level Low-level

Ml Audio:

Title(Attribute) Fight Club (value) laughter(attr.) - 0%(value)

Genre Drama/Thriller/Myst./ perc. of speech - 60%

Cast Edward Norton, Brad audio loudness - 8/10

Director David Fincher music - 40%

Time 139 Min ...

Producer Art Linson, Ross Bell

Date 12/05/2015 Video:

Broadcaster 20th Century Fox Motion activity - 7

Language English ...

Country USA

Year of production 1999

Certification Argentina: 18, Textual:

Ratings 8,5/10 Language - English

Formal - 20%

Informal - 80%

Scene: indoor - 60% color - RGB = (123:125:123) luminance - 30% chrominance - 40% edge features

Mood: anger 70% depressed 90%

As can be seen from this table both the high-level and low- level features have attribute and attribute values. Ml relates to movie 1 and Ml relates to movie 2 and so forth.

Recorded data set A recorded data set is preferably content data that a user wants to enjoy. For example it may be a movie that a user wants to see, in a near or distant future. The user is able to select one or more data set comprising content data, which then becomes recorded data sets. These recorded data sets may have a fixed delivery date and time, whereby a user has to be near the electronic device in order to enjoy the content. In another embodiment the recorded data set may not have a fixed delivery time. In this cases a user is able to select a date and time when he/she wants to enjoy the content data. The user may also be able to define a time and date window wherein he/she wants to enjoy the content data.

This makes it possible for the present invention to let two or more users to enjoy the content data at the same time so that they can discuss what they have experienced afterwards.

For example the system may match the date and time between users that want to enjoy the same content data. This may also be possible between users living on different locations having a time difference, since the system may be able to calculate the real viewing date and time. In this way a person in Japan and a person in the USA are able to watch a movie together.

Furthermore the system preferably only invites user profiles wherein the other matching criteria regarding personal profile, partner profile, viewing profile, like list, dislike list and so forth is meet, according to the thresholds.

Match rate

Match rates may be calculated by representing the viewing profiles as vectors with values, subtracting the vectors and adding or multiplying the values of the differential vector. Like rate / dislike rate

The like rate and dislike rate may be calculated by comparing number of data sets that are the same and divide that value with the total number of data sets.

For example a first user profile may have three movies that are the same as in a second user profile. If the total number of movies in the second user profile is 5 the like rate could be calculated by dividing 3 with 5, thus the like rate is 60%. However the number of movies may also be divided by the total number of movies in the first user profile in order to end up at a like rate.

The above example is only one way to calculate the like/dislike rate, other methods to calculate the like/dislike rate may be applied and are obvious to the person skilled in the art.

Example of scenarios wherein the present invention may be used.

Below follows two scenarios wherein the present invention may be used with great advantage.

Scenario 1 - dating scenario Let say that a person A is a hardworking engineer and during periods of a year he/she spends a lot of time at work. Person A thinks that one of the best way to relax is to watch a movie after work or on weekends. However person A is also very social and thus is keen on meeting new people and make new contacts, maybe also meet a potential partner since person A at the present time is single. Person A has recently bought company X's top-of-the-line-unit (electronic device) for viewing movies. The unit can download content data such as movies from the Internet and thus, A does not have to go to the video store for renting movies.

When A connects the unit to the TV and Internet the unit is equipped with hardware and software that helps A to set-up and install the unit. Upon installation the person A may be asked if he/she wants to join an online dating service for meeting other people via the unit. Person A chooses this option and is directed to a page for creating a user account. The set-up page for the user account preferably asks person A about his/her own characteristics such as age, height, hair color, eye color, language, location, income, profession, personality and so forth. The questions that person A does not want to answer is replaced with a default value.

Next step may be to ask person A about desired characteristics that a potential partner B preferably should have. A similar questionnaire is presented to person A to fill in and the boxes not answered may be filled with a default value. Now the active set-up for person A is done and no more time needs to be spent on this task, unless A wants to change the settings.

Thereafter person A starts to view movies on the new unit and preferably every time a movie is finished a window pops up asking person A if he/she liked the movie or not. Preferably there is only two choices yes or no, however in some circumstances a 5 or 10 degree Likert scale may be presented.

Preferably this question is asked right after the end of the movie, in this way persons A first spontaneous reaction provides the answer to the question. If too long time passes the person may talk to other people and thus the answer may be influenced by other people's opinion. When A selects yes or no a number of things happens. The movie just seen is preferably stored in a like or dislike list associated with person As user profile. The like and dislike list is also used for finding and matching potential partners on the dating service that person A joined in.

Thus if As personal profile and partner profile is matched with another persons personal profile and partner profile so that the match rate exceeds a certain threshold such as 50%, 60%, 70%, 80%, 90% or 99%. The like list and/or dislike list in the user profile is used for measuring a match rate for the two persons. If this match rate also reaches a certain predetermined threshold such as 50%, 60%, 70%, 80%, 90% or 99%, a matching message is created and is sent to both person A and the other person. The matching message may comprise a photo, and description of the other person derived from the user profile. However the matching message preferably does not comprise real contact information for security reasons. The persons may contact each other via the dating service without revealing their real contact information. In this simplest embodiment of the dating service the persons may be matched by the name of the movies or genre of the movies, thus by high-level features.

However, when the person A has watched a movie and added it to his/her like list further steps may be taken in order to analyze the movie on a deeper level. For example high-level features are extracted from metadata associated with the movie. Preferably the high-level features relates to title, cast, genre as described earlier in this document.

Furthermore low-level features are extracted from the data set representing the movie, in order to analyze what type of movie personality person A actually is. Preferably low- level features relates to audio, video, textual, scene, mood, human, and camera motions features as described earlier in this document.

When the high-level and low-level features are extracted and stored in the user profile matches may be carried out again, now considering an even deeper view of the user profile. Now again the high-level features of person A may be compared to the high- level features of another persons' user profile and if the match rate exceeds a predetermined threshold such as 20, 30%, 40%, 50%, 60%, 70%, 80%, 90% or 99%, the system may either send a matching message or preferably it may continue to the next step wherein the low- level features are matched. The low- level features of the media watched by person A is preferably also compared to low-level features of the other persons user profile and if the match rate exceeds a predetermined threshold such as 50%, 60%, 70%, 80%, 90% or 99% the system may send a matching message.

Preferably when both the high-level and low- level features exceed a certain predetermined threshold a message may be sent to the matched persons.

Preferably the high-level features and low- level features in the like list of person A are matched with the high-level features and low- level features in the like list of an other person. The high-level features and low- level features in the dislike list of person A are preferably matched with the high-level features and low- level features in the dislike list of an other person.

In another embodiment a certain combination of the match rates from the high-level and low- level matches may trigger the system to send a matching message. The predetermined threshold may be set automatically based on statistical analyses and earlier experience. However the predetermined thresholds may also be set by the holder of the user profile, thus in this example by person A.

In a further embodiment of the present invention person A may want to watch a movie together with another person. Since the newly installed unit provides the possibility of downloading movies over the Internet it is possible for person A to select a movie and a viewing date and time. Furthermore person A may also mark a check box that the dating service should find a matching user profile that wants to watch the same movie at the same time. Or the dating service may automatically send an invitation message to A and another person having a matching user profile and propose to watch the movie together.

In this way a new contact between two persons may be created and since they have watched the same movie it may be easier to find conversation topics, instead of trying the usual "ice-breakers".

In the above scenario instead of a movie the content data may relates to TV- programs, radio programs, computer games, web pages and so forth.

For example people may be invited to join the same team in an online computer game or antagonists may be invited to play against each other.

In an embodiment related to online computer games, the user profile may further comprise a game profile (GP) comprising characteristics associated to games such as, aggressiveness, speed (slow, fast), risks (risk avert, risk taker), success rate, failure rate, number of kills, number of times being killed, stealth, preferred weapon, and so forth.

In this way, teams for online gaming can be created and so forth.

Scenario 2 - person matching job position scenario

In a second embodiment a person A may be a job seeker and an organization may have open job positions, or an organization may look for a certain personality for a position within the company.

In this scenario a person creates a user profile according to above, however in this case the user profile is preferably more focused on knowledge, profession, education, experience and so forth. However personality characteristics are also important. An entity of the company, for example the human resource department or another department creates the company user profile and also the job opening profile (partner profile) that meets the demand for the company and the open job position. For example important personality characteristics for the job position is filled in, experience, profession, language, cultural experience and so forth. When this system is installed on a company network, each user computer may be equipped with software that measures the high-level features and low- level features of the applications and/or WebPages an employee is using or visiting. Furthermore each computer may be equipped with a special application so that an employee can personalize his/her computer, each option of the application may be related to certain personality profile.

Furthermore the software may count number of time the computer have been used for presentation, analyze the Power point pictures, sound used? what sounds? colors, number of slides text, language and so forth. Thus analyzing low- level features as described earlier in this document and associate the low- level features with the user profile. Furthermore the system may ask each employee to fill in the user profile regarding personal profile and job opening profile (partner profile).

Accordingly the questions are therefore aimed at working conditions and also personality for a certain job position. For example a sales person may have a different personality compared to persons working with product development. This embodiment may not only be for internal job openings within a company, other job seeking services on the Internet can also use this service.

Detailed description of how to build and use the invention.

High-level View

The main idea is to match users based on their viewing history and some demographic data. The use of viewing profiles guarantees that the users cannot cheat. This can possibly happen unwillingly in a dating service currently when filling out questions regarding themselves. When a matching partner is found the system can propose the two users to meet on an online chat during the broadcast of a program they both like or when watching a program they have both recorded on their personal video recorders. Algorithm

There are two possible embodiments of the online matching system. One is empirical which is based on our knowledge on how things should work. The other is grouping based on a priori knowledge about couples that are together.

Empirical method: The system collects viewing preferences and history (user profile), detects clusters of programs users like/dislike and searches for matching partners. This search is performed in a subset of the users that match other criteria explicitly indicated by the users personal profile and partner profile. The matching is performed in such a way that two user profiles are considered for a date preferably if the intersection between the like/dislike groups of the potential partners is large but not complete. A complete intersection would probably result in two persons with very similar interests and tastes thus probably not interesting for a date. A large but not complete intersection of tastes and interests can on the contrary be considered interesting. The threshold of likeliness may be selected by the user and shown when proposing the date.

A priori knowledge method:

This method preferably considers viewing profiles of couples that have dated in the past and are together currently. The method may find the amount and type of overlap (or patterns) in the programs/movies that they viewed in the past. For example a person who watches news may be best coupled with a person who watches sitcoms etc. Or the grouping may be at the program level if the people stay in the same country. So people who watch "Star Trek" have a common interest. Once enough patterns are accumulated, then, given a set of viewing profiles, along with some personal data (personal profile, partner profile) the method can find suitable matches for a user.

For people from different countries, the method may have to analyze at the deep level and figure out concepts and also their look and feel rather than the program titles and their description. So the method may find that people who watch programs that have a high cut rate like people who watch programs with a similar cut rate; people who watch emotional movies like people who watch sci-fi programs etc.

Online Viewing Date

If a user wants to watch a recorded program, the system could search for matching profiles of persons that have recorded the same program and send an invitation for watching the program together. This would create a social atmosphere of sharing an experience with a group that is sometimes present when watching live TV but is not present when watching a recorded program.

Associating high level person information with viewing patterns

Figure 3 gives an overview of the process. We take in viewing histories (VH) of the recorded programs. The viewing histories preferably contain all the Electronic

Program Guide (EPG) high-level information about the programs such as title, genre, cast, director, time, date.

For the selected programs, the system extracts low- level features (step 1) (such as shot cuts, motion activity, dominant colors, etc.) from the corresponding video streams (VS). This can be either in real time when the user is watching TV, or for the stored programs.

Next (step 2), programs may be grouped based on the extracted features. These groups preferably have labels associated with the low-level features. For example, if color information is used then the groups will have labels: dark vs. light, warm colors vs. cold colors.

Next (step 3), it is possible to derive combined labels based on the high-level information, plus the low feature labels. For example, we can say that a particular group is "comedy," "dark" and "slow". As a result of this process, now we have personal genres that are much more descriptive than just the viewing history comprising EPG high-level information.

Next step (step 4) is to derive a mapping between the personal profile and the personal genres for this user.

Finally (step 5), mapping tables may be created. Association of personal profile and personal genres may be performed according to the flowchart in figure 3. Wherein, VH= Viewing history, VS= Video stream and PePH=Personal Profile.

The personal profile may be given as:

Computer system

Referring to Figure 4, a computer system according to the present invention may comprise a database DB for storing user profiles, an electronic device (11) for receiving/downloading/sending/amending information such as user profiles, a network 12 such as the Internet, a first user profile 13 and a second user profile 14. Furthermore the electronic device may be any other kind of electronic device such as mobile phone, PDA, game console, DVD, and so forth. Furthermore the computer system may comprise a database for storage of content data (15), and a processing unit (16) for extraction of high-level and low- level features.

Extraction of features may also be performed on the electronic devices (11). Matching People One problem to solve is how to match people based on their personal profile (PeP) versus a potential partner profile (PaP). The users themselves may decide how much overlap should be there. However in another embodiment this may be automatically calculated by the use of statistics and earlier experience of the system. The personal and partner profile matching is based on people's preferences (e.g. American guy searching for a Korean lady of specified age).

The high-level feature matching is preferably based on facts about the content data, e.g. Title, genre, cast, production year and so forth. Preferably once the profile matching and high-level feature matching is done, the true personal attributes, the low- level features may be matched.

Different factors are important for relationship quality such as personality and mental health, as well as self-monitoring. Self-monitoring means whether a person acts differently in different contexts (high self-monitoring) or a person who acts the same no matter what context (low self-monitoring). Research shows that individuals who match up with similar self-monitoring scores have a tendency to become partners romantically, although the quality of the relationship does depend on which type of self-monitor the individuals are. Less involvement and less commitment are seen with dyads of high self- monitors, whereas characteristically low self-monitors appear more committed. The system may need input on how consistent users are with their personality across all situations. The system might be more successful in matching people with consistent personality than those whose personality is constantly changing.

Applications of the invention

The invention may be applied to personal video recording systems and services but also to video on demand systems, mobile devices, game consoles and so forth.

The same idea can be applied to audio systems, online music services, online gaming services, job matching systems, Internet radio services and so forth.

It will be apparent to a person skilled in the art that the invention can also be provided in the embodiment of a computer programme product comprising computer executable code for programming at least one processor 502 to execute the method according to the invention on embodiments thereof.

This computer programme product may be stored on a carrier at a home location of a user, like a local hard disk drive 501 in a personal computer 504 or optical disk 503, or on a remote server 505 of a service provider using the method in a return for a fee. In return for the fee, the service is delivered over a network 506 to a client application 502 with a user.

In the above description the term "comprising" does not exclude other elements or steps and "a" or "an" does not exclude a plurality. Furthermore the terms "include" and "contain" does not exclude other elements or steps.

Claims

CLAIMS:

1. A method of providing matches on an online dating system, said system comprising: a first user profile, and at least one second user profile, the profiles each further comprising: a personal profile comprising at least one personal property comprising a value, a partner profile comprising at least one partner property comprising a value and - a viewing profile, the method comprising steps of: comparing the partner property values in the partner profile in the first user - profile with the personal property values in the personal profile in the second user profile, comparing the partner property values in the partner profile in the second user profile with the personal property value in the personal profile in the first user profile, establishing a match when, upon comparison, at least a pre-determined number of the properties of the first and second user profile have substantially same values, comparing the viewing profile in the first user profile with the viewing profile in the second user profile in order to calculate a match rate, and - when the match rate exceeds a predetermined threshold, providing at least one of the first user profile and second user profile with a matching message.

2. A method according to claim 1 wherein the viewing profile further comprises: a like list, and - a dislike list, the like list and dislike list comprising data sets , the method further comprises steps of: comparing data sets in the like list in the first user profile with data sets in the like list in the second user profile to calculate a like rate, comparing data sets in the dislike list in the first user profile with data sets in the dislike list in the second user profile in order to calculate a dislike rate, comparing at least one of the like rate and/or dislike rate to at least one predetermined threshold, and - if at least one of the like rate and/or dislike rate exceeds the predetermined thresholds, sending matching messages to the first user profile and second user profile.

3. A method according to claim 2 wherein the data sets further comprising high- level features and low-level features, the method further comprises steps of: - extracting high-level features from the data sets in at least one of the like list and/or dislike list, extracting low- level features from the data sets in at least one of the like list and/or dislike list, creating a like group comprising high-level features and low-level features from the like list, creating a dislike group comprising high-level features and low-level features from the dislike list, comparing the like and dislike groups of the first user profile with like and dislike groups in one or more second user profiles in order to calculate at least one feature rate, and comparing the feature rate with at least one predetermined feature threshold, and upon approval, sending matching messages to the first user profile and second user profile.

4. A method according to claim 3 wherein the high-level features are chosen from a group of features comprising: title, genre, - cast, director, time, date, broadcaster, language, country, year of production, certification, and - ratings.

5. A method according to claim 3 wherein the low- level features are chosen from a group of features comprising: audio features, - video features, textual features, scene features, mood tone features, human features, and - camera motion features.

6. A method according to claim 3, wherein the high-level features and low- level features are computed from the audiovisual signal by content analysis algorithms.

7. A method according to claim 1, wherein the online dating system further comprises a database comprising couple histories, the couple histories comprising matched user profiles that are still together, the method comprises steps of: from the matched user profiles, extracting matching patterns relating the matched user profiles, - storing the extracted patterns in the database, creating matching rules from the stored patterns, and using the matching rules for comparing a first user profile with one or more second user profiles.

8. A method according to claim 1, wherein the user profiles further comprise at least one recorded data set, the method comprising steps of: finding second user profiles comprising the same recorded data set as the first user profile, sending a viewing proposal to the first user profile and at least one second user profile having the same recorded program, so that the profiles can view the recorded program at the same time.

9. A method of creating a mapping table in an online dating system comprising: - a user profile, at least one viewing history, and at least one recorded data set, the at least one viewing history comprising (EPG) high-level features from the recorded data sets, the user profile further comprising: at least a viewing profile, the method comprising steps of: extracting low- level features from the recorded data sets, grouping the data sets, based on the low-level features, - assigning each group one or more label(s) associated with the low-level feature(s), based on the low-level features and the high-level features, deriving combined labels related to a group so that the group is defined by the combined labels, matching the viewing profile in the user profile with the combined labels in order to map suitable data sets, and storing these matches in a mapping table.

10. A data structure comprising: a user profile, - the user profile comprising: a personal profile, a partner profile and a viewing profile.

11. A data structure according to claim 10 wherein at least one of the personal profile and partner profile comprises at least one of the following properties: name, gender, age, nationality occupation, income range, ethnic background, - sports, arts, personal attributes, and personality traits.

12. A data structure according to claim 10 wherein the viewing profile further comprises at least one of a like list and a dislike list.

13. A data structure according to claim 12 wherein at least one of the like list and dislike list further comprises at least one of high-level features and low- level features.

14. A computer system comprising: a server for storing at least one or more user profiles in a database, the server being connected to a network, an electronic device for downloading data sets and for downloading at least parts of at least one of the one or more user profiles from the database via the network, each user profile comprising: a personal profile, a partner profile, and a viewing profile, said computer system being programmed to: compare a first user profile with one or more second user profiles based on at least one of the personal profile and the partner profile and the viewing profile, upon match, sending a matching message to at least one of the first user profile and second user profiles.

15. A computer system according to claim 14 further being programmed to compare user profiles according to claim 1.

16. Software on a data carrier executable on computing hardware for executing the method according to claim 1.