WO2016203179A1

WO2016203179A1 - Method and system for recommending content to a user based on the user's interaction history

Info

Publication number: WO2016203179A1
Application number: PCT/FR2016/051488
Authority: WO
Inventors: Paul DE MONCHY; Gabriel MANDELBAUM; Thibault D'ORSO
Original assignee: Spideo
Priority date: 2015-06-18
Filing date: 2016-06-17
Publication date: 2016-12-22
Also published as: FR3037758A1; FR3037758B1

Abstract

The invention relates to a method and system for recommending content to a user. According to the invention, first parameters (T3, T4, T5, T8), known as themes, and/or second parameters, known as moods (M3), are assigned to content. Interactions (WTCH) between the user and a content item are identified and, based on the interaction and the first content item (C1), a set is generated, known as the user's semantic footprint, comprising pairs (P1, P2, P3, P4, P20) of mood(s) and/or theme(s) defined according to the theme(s) and/or mood(s) of the first content item (C1). Each of the pairs (P1, P2, P3, P20) in the user's semantic footprint is assigned a weighting (W_P1, W_P2, W_P3, W_P4, W_P2O) defined based on the first interaction (WTCH). The content is then prioritised according to the weightings (W_P1, W_P2, W_P3, W_P4, W_P2O) of the pairs (P1, P2, P3, P20) in the user's semantic footprint and, subsequently, according to content score within a pair, such that content is recommended without the user having to select a given mood or a theme. Such a recommendation method and such a recommendation system, also known as Profile-Based Recommendations, allows personalised content recommendations to be made based on a semantic intensity network.

Description

METHOD AND SYSTEM FOR RECOMMENDING CONTENT TO A USER BASED ON THE INTERACTION HISTORY OF THE USER

FIELD OF THE INVENTION

The present invention generally relates to methods and systems for recommending content.

PRIOR ART

We see that the majority of users who turn on their television do not know which program they are going to watch. In a context where consumers of audiovisual content have access to dozens of television channels and thousands of on-demand or catch-up films or programs at all times and on all their screens, it is useful for the user can benefit from a help with the discovery and the recommendation of contents to be found in the plethoric offer of contents.

However, it is important for the satisfaction of the user to be offered a selection of relevant content.

Video hosting platforms are known from the state of the art and also offer a solution for recommending content. According to this recommendation solution, the contents are associated with simple labels, called tags, corresponding to particular themes or genres.

The recommendation of contents is made on the basis of the tags defined by the person who posted the video and the analysis of similar uses of this video among the users. This type of recommendation, also known as "collaborative filtering", is not fully satisfactory because (i) the tags applied by the different posters may not be well adapted to the content, (ii) the filtering operated by the analysis of uses of similar profiles does not make it possible to exploit the diversity of contents of a catalog, (iii) the system is dependent on the analysis of the interactions of a very large number of users, technically difficult to implement in real time, to achieve an acceptable level of relevance. Such a recommendation solution will tend to recommend to all users the most popular videos among videos associated with a small common denominator of frequently assigned tags.

US20120324374 discloses a film recommendation system which provides for assigning weighted micro-categories to movies, presenting multiple micro-categories to the user to actively declare them as "convergent" or "divergent", and define user profiles including micro-categories and weights associated with these microcategories. However, the system of document US20120324374 requires the user to declare preferences (the micro-categories presented to him for selection), which is restrictive and not very fluid from the point of view of the user experience. Above all, the fact that the user has to declare, that is to say select or define, micro-categories makes the system deterministic, that is to say constrained by the specific declarations of the user to a given moment, so that the resulting recommendation does not allow to follow real-time evolution of the user's tastes.

The object of the present invention is to propose a novel method and system for recommending content that makes it possible to overcome all or some of the problems set out above.

In particular, the present invention aims to propose a new method and content recommendation system for providing personalized recommendations, varied and relevant, taking into account in real time, the content or contents with which (s) the user has already interacted.

SUMMARY OF THE INVENTION For this purpose, the subject of the invention is a method for recommending content to a user, implemented using a computer system, for each of the contents being attributed to:

□ first parameters of themes, called themes, and

□ second mood parameters, called moods, each theme being associated with one or more moods;

said method comprising the following steps:

- identifying an interaction, called the first interaction, of the user with one of the contents, called first content;

generating, as a function of said first interaction and said first content, a set, called the user's semantic fingerprint, which includes pairs of mood (s) and / or theme (s) defined according to the themes ( s) and mood (s) of said first content;

determination for each of said pairs of the semantic imprint of the user of a weighting defined as a function of said first interaction;

- identifying a second user interaction with content, called second content; and

□ a generation step in the semantic fingerprint of the user, at least one new pair corresponding to a pair of theme (s) and / or mood (s) present in the second content, and a step of determining the corresponding weighting of the new pair as a function of said second interaction;

□ and / or, a step of modifying, as a function of the second interaction with the second content, the weighting of at least one pair of said user's semantic imprint,

- determining a value, called score, for each content;

- scheduling of contents by putting first and foremost the contents which present a pair of theme (s) and / or mood (s) corresponding to the pair of the semantic fingerprint of the user of greater weighting, compared to other contents presenting a pair of theme (s) and / or mood (s) corresponding to a pair of the semantic imprint of the user of lower weighting,

the contents associated with the same pair being ordered together according to their score,

so that the contents are recommended without the user having to select a theme or mood.

Thus, the weighting of the elements of the semantic imprint of the user is a function of the interaction history of the user, and the contents are ordered in priority according to the weightings of the pairs of theme (s) and / or mood (s) of the semantic imprint of the user and the presence or not) of these pairs in the contents.

Once the contents are prioritized according to the weighting levels of the pairs of the user's fingerprint, scheduling continues within each of these pairs. In particular, the contents that correspond to a given pair of the semantic fingerprint of the user are ordered among themselves according to their score. The score of a content can be impacted as detailed below by a weighting vector.

The scheduling of the contents with respect to the different pairs of the semantic fingerprint of the user is performed so as to recommend contents of a given weighting pair in front of contents of a lower weighting pair. In other words, the recommendation method gives priority to contents containing themes and / or moods corresponding to a pair of strong weightings present in the user's footprint in relation to the score of the contents which is a secondary criterion of scheduling.

The score of a content can be determined by any score calculation mechanism. For example, it can be expected that the score is calculated using a statistical calculation mechanism, such as a score calculation by popularity index. Advantageously, the same mechanism is used to calculate the scores of the different contents. The contents can be recommended to the user by displaying them in descending order of their scheduling. In other words, the scheduling makes it possible to prioritize the most relevant contents with regard to the user's semantic imprint and thus to recommend them to the user independently of or by limiting the impact of the analysis of the user. use of other users.

The construction of the semantic imprint of the user is done transparently for the user, that is to say automatically without the user having to perform particular steps via a dedicated construction interface. In particular, the user does not need to declare, that is to say to inform or select, particular parameters of contents by means of a specific interface so that the user imprint is constructed, he it is enough to interact with the content (eg reading, notation, ...). In other words, the construction of the impression does not require the user to modify his behavior. Similarly, it can be provided that the contents ordered according to the semantic fingerprint of the user are automatically displayed to the user, for example on a computer system screen which implements the recommendation method, for example a computer fixed or portable, a tablet, a smart phone or a smart TV. This recommendation can be implemented automatically when the user turns on the system.

Such a method and recommendation system, also called Profile-Based Recommendations, make it possible to make a personalized recommendation of content based on a user profile formed by the semantic imprint of the user and a network. of semantic intensity formed by the interaction of different themes and moods and their impact on the scheduling of content. Indeed, the themes and / or moods, and possibly the weighting vectors as detailed below, associated with the contents and found in the user's footprint form a network or graph of semantic parameters that make it possible to effectively recommend of the content according to the preferences or desires of the user.

Realizing a user profile using pairs of two distinct types of parameters, one relating to a theme that "objectively" characterizes one content, and the other relating to an emotion that "subjectively" characterizes what can feel the user, provides specificity and reliability in the recommendation of content vis-à-vis mood and interests of the user.

Compared to the known solutions of recommendation, the method of recommendation according to the invention does not consist simply of taking into account a second parameter juxtaposed or completely independent of the first parameter, but on the contrary of combining these parameters in their attribution to the contents and in the construction of the semantic imprint, with a very specific relationship between these two parameters. Indeed, each theme is associated with one or more moods and furthermore as explained below the weighting of a mood in a given content is a function of the themes that are associated with this mood and attributed to said content.

This combination relationship between the two types of parameters assigned to the contents and used to construct the semantic imprint makes it possible to obtain a reliability of recommendation results with regard to the mood and interests of the user.

The pairs are built with cross-over settings that are both emotion and user's interests. It is thus possible to obtain a thematic and emotional relevance of the contents proposed to the user.

According to one embodiment, said pairs of the imprint are generated from the themes and / or moods of the contents with which the user has interacted over a predefined time.

Following the generation of new pairs and / or the modification of weighting of pairs, the pair of strongest weighting is determined and the contents associated with this pair are ordered in a decreasing manner according to their score.

Said first and second interactions may be of the same type or not, for example a second viewing or a notation. The second content may be the same as the first content in the case of different interactions.

According to one characteristic of the invention,

an additional parameter, called the weighting vector, being assigned to at least one of the contents with which the user interacted, said weighting vector targeting a given theme or mood,

then the method comprises the step of associating said weighting vector with the pairs of the fingerprint, called vectorized pairs, other than those comprising the targeted theme, which are generated in the semantic fingerprint on the basis of said content,

and the score of a content which comprises one of said vectorized pairs, is determined according to the presence or not in said content of the theme or the mood targeted by the weighting vector.

In other words, the score of a content relative to a given pair of the semantic imprint can be impacted by the presence or absence of a theme or mood in said content, when a weighting vector is associated with said pair.

In particular, the content score is reduced when the target theme or mood is not assigned to said content.

The use of weighting vector makes it possible to reduce or increase the influence of a pair in one or the other of the recommended contents within a pair according to the presence of a mood or a theme. given in said content, which makes it possible to refine their order with regard to the same pair. The weighting vector can potentially be associated with any pair of theme (s) and / or mood (s) resulting from the semantic fingerprint of a content belonging to the user's interaction history. .

Advantageously, the weighting vectors are assigned to some of the contents at the same time as the other mood and theme parameters.

According to one characteristic of the invention,

a weighting being assigned to each mood of a content according to the themes associated with said mood and attributed to said content, and

a mood of a content that presents a weighting value beyond a given threshold value, being called the main mood,

the weighting determining step for each pair of the imprint comprises assigning a markup to the weighting of a pair of the semantic imprint of the user, when one of the moods composing said pair is a main mood, or when one of the themes composing the pair is associated with a main mood.

In particular, the impact of the interaction on the weighting of said pair is increased.

According to one characteristic of the invention,

the generation of a pair of the semantic imprint of the user from a mood or two moods present in a content with which the user has interacted, is performed only if the or each of said moods is a main mood.

In other words, the generation of a new pair comprising a mood is thus achieved if the mood is a main mood, that is to say a mood whose weighting is greater than a threshold value.

In the case where the second content is different from the first content and where the theme (s) and / or mood (s) differ between the two contents, the difference between two moods of the same designation may relate to the main or secondary character of the mood. because of a difference in weighting. Thus a mood can be considered as the main mood for a content and as a secondary mood for another content.

However, it can be expected that a pair includes a theme associated with a secondary mood.

The fact that a pair of the semantic fingerprint of the user is found with a greater weighting than the other pairs, can result from a filtering operation, for example according to temporal data, excluding some more weighting pairs. important. Thus, the content recommendation based on the highest weighting pair continues with the next highest score pair remaining after possible exclusion of certain pairs.

According to one characteristic of the invention, said method comprises a pair recommendation exclusion step according to which the recommendation of content (s) associated with a first pair of the semantic imprint of the user, leads to exclusion of a second pair from the semantic imprint of the user who presents a common theme or mood with a theme or mood of the first pair of the semantic imprint of the user.

According to a feature of the invention, said method comprises a content exclusion step relative to a pair of the semantic fingerprint of the user, according to which the content recommendation (s) associated with a first pair of the semantic imprint of the user leads to the exclusion of this (these) same content (s) in a second pair of the semantic imprint of the user to whom the user (s) is or are also associated (s) content (s). It can be expected that when content is recommended with respect to a given pair of the user's fingerprint, these same contents are then excluded from the other pairs for the rest of the recommendation.

According to a particular aspect, the pairs of the semantic imprint according to which the contents are ordered, are filtered according to the number of contents corresponding to these pairs, the number of contents being possibly impacted by the exclusion of content as mentioned below. above. Thus, if, after recommending the content of a first pair, contents are excluded from a second pair and the number of contents remaining associated with the second pair is less than a threshold number, said second pair is itself excluded from the recommendation method. .

According to one characteristic of the invention, the weighting of a pair of the semantic fingerprint of the user, generated as a function of an interaction with a content, is a function of the type of the identified interaction.

According to a characteristic of the invention, said method comprises storing filter data, for example temporal data, such as a consultation time and / or the date and / or the time of the implementation of said method of recommendation, according to which the contents are ordered. The pairs of the fingerprint can be filtered according to said filter data.

According to one characteristic of the invention, the interaction of the user with at least one of the contents comprises:

- the consultation of information, for example a card, relating to said content; and or

reading of said content, for example for viewing and / or listening; and or

- adding to a reading list; and or

- alerting said content; and or

- the recording of said content; and or

- the notation of said content. The invention also relates to a computer system for recommending to a content user, with each of the contents being attributed:

□ first parameters, called themes, and

□ second parameters, called moods,

each theme being associated with one or more moods

said system comprising:

means for identifying an interaction, called the first interaction, with a content, called the first content, with which the user has interacted;

generation means, as a function of said first interaction and said first content, of a set, called the user's semantic fingerprint, which includes pairs of mood (s) and / or theme (s) defined as a function of the themes and moods of said first content;

determination means for each of said pairs of the imprint of a weighting defined as a function of said first interaction;

means for identifying a second interaction of the user with a content, called second content; and

□ generating means in the semantic imprint of the user, at least one new pair corresponding to a pair of theme (s) and / or mood (s) present in the second content, and a step of determining the corresponding weighting of the new pair as a function of said second interaction;

□ and / or modification means, as a function of the second interaction with the second content, of the weighting of at least one pair of said user's semantic imprint,

means for determining a value, called score, for each content;

means for scheduling the contents by placing first and foremost the contents which present a pair of theme (s) and / or mood (s) corresponding to the pair of the semantic fingerprint of the user of greater weighting, by relative to other contents presenting a pair of theme (s) and / or mood (s) corresponding to a pair of the semantic fingerprint of the lower weighting user, said scheduling means being configured also to order between them the contents associated with the same pair according to their score, so that the contents are recommended without the user having to select a theme or a given mood.

According to an advantageous characteristic of the invention,

an additional parameter, called the weighting vector, being assigned to at least one of the contents with which the user interacted, said weighting vector targeting a given theme or mood, the system comprises means for associating said weighting vector the pairs of the fingerprint, called vectorized pairs, other than those comprising the targeted theme, which are generated in the semantic fingerprint on the basis of said content,

and, for a content that includes one of said vectorized pairs, said score determining means is configured to determine the score of said content as a function of the presence or absence of said content of the theme and / or of the mood targeted by the vector of weighting.

According to an advantageous characteristic of the invention,

the weight determination means for each pair of the imprint are configured to assign a markup to the weighting of a pair of the semantic imprint of the user, when one of the moods composing said pair is a main mood, or when one of the themes composing said pair is associated with a main mood.

According to an advantageous characteristic of the invention,

a weighting being assigned to each mood of a content according to the themes associated with that mood and attributed to said content,

- and a mood of a content that has a weighting value beyond of a given threshold value, being called the main mood, the pair generation means are configured such that the generation of a pair of the semantic fingerprint of the user from a mood or two moods present (s) in a content with which the user has interacted, is realized only if the or each of said moods corresponds to a main mood.

The invention also relates to a computer program comprising program code instructions for executing the steps of a method as described above when said program is executed by a processor forming part of a computer system. such as a computer, smart TV, tablet or smart phone.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention will be better understood on reading the following description of exemplary embodiments, with reference to the appended drawings in which:

FIG. 1 is a schematic representation of moods and themes associated with a first content C1, according to one embodiment of the invention;

FIG. 2 is a schematic representation of the semantic fingerprint associated with a user after viewing by the user of the first content C1;

FIG. 3 is a schematic representation of moods and themes associated with a second content C2, according to one embodiment of the invention;

FIG. 4 is a diagrammatic representation of the semantic fingerprint derived from that of FIG. 2, but impacted after a notation by the user of the second content C2;

FIG. 5 is a schematic representation of three other contents which each present the elements of the pair of the semantic imprint of greater weighting;

FIG. 6 is a schematic representation of the ranking of the three contents C3, C4, C5 of FIG. 5 according to their score, the score of one of the contents being impacted by a weighting vector associated with said largest weighting pair;

FIG. 7 is a schematic representation of the ranking of two other contents C6, C7 according to their score, these two other contents being associated with the highest weighting pair which remains after the recommendation of the contents of the pair of FIG. ;

Fig. 8 is a block diagram illustrating steps of a content recommendation method according to an embodiment of the invention.

DETAILED DESCRIPTION

The concept of the invention is described more fully hereinafter with reference to the accompanying drawings, in which embodiments of the concept of the invention are shown. However, this concept of the invention can be implemented in many different forms and should not be construed as limited to the embodiments set forth herein. Instead, these embodiments are provided so that this description is complete, and communicate the scope of the concept of the invention to those skilled in the art. The scope of the invention is therefore defined by the appended claims.

Reference throughout the description to "an embodiment" means that a particular feature, structure, or feature described in connection with an embodiment is included in at least one embodiment of the present invention. Thus, the appearance of the phrase "in one embodiment" at various locations throughout the specification does not necessarily refer to the same embodiment. In addition, the particular features, structures, or features may be combined in any suitable manner in one or more embodiments.

Referring to the figures and as recalled above, the invention relates to a method and a system for recommending to a content user. according to an embodiment, said contents are multimedia-type contents, such as films, television programs or videos. Content may also be books, music, or other types of content.

In the remainder of the description, said contents are films. But of course the description is applicable to other types of content.

It is intended to assign to each content C1, C2, C3, C4, C5, C6, C7 first parameters T1, T2, T3, T4, ..., T30 called themes (step 1 10). Second parameters M1 to M9 called moods, are also assigned to the contents C1, C2, ..., C6, C7. Said awarding step may be part of the recommendation method or external to said method by being performed previously, but said awarding step is not performed by the user to whom content will be recommended. Said allocation step may be performed using a specific man-machine interface, by an expert operator who is distinct from the user. The associations between the contents, the theme parameters and moods, and the weighting vectors (presented below) are stored in a database, called the content database. The content database can be part of the system or is accessible to the system.

Each theme is by default associated with one or more moods. Several themes can be associated with the same mood and a theme can be associated with several moods. A weighting is also assigned to each mood relative to the content to which said mood is assigned.

The weighting of a mood in a given content is determined according to the themes that are associated with this mood and attributed to said content. The weighting of a mood can be defined by applying a mathematical formula or using a table of correspondence.

The weighting of a mood can vary for the same mood from one content to another. According to a particular aspect, said themes and mood attributed to a content are derived from a set of themes and moods defined so that each theme of said set of themes is associated with at least one mood. In other words, each theme carries weighting in one or more moods, so that when a theme is assigned to a content, the mood or at least some of the moods in which said theme is weighting is also assigned audit content. Thus when a theme is assigned to a content at least one of the moods with which it is associated is assigned to said content. It can also be expected that a mood is assigned to a content without a theme that is associated with this mood is assigned to said content.

All of the theme (s) and / or mood (s) assigned to a content as well as their weight, thus form, possibly with other parameters such as a weighting vector, the semantic imprint of the content.

As illustrated in FIG. 1, the content C1 that corresponds to the movie "Inception" includes the following themes and moods:

- the mood M1: "Twisty" whose weighting WMI is equal to 10, with the themes T1: "Masterplan" and T2: "Intricate narrative";

the M2 mood: "Spectacular" whose weighting WM2 is equal to 9;

- the mood M3: "Suspense" whose weighting WM3 is equal to 5.5 with the theme T3: "Beat the dock";

- the mood M4: "Strange" whose weighting WM4 is equal to 7.5, with the themes T4: "Dreamscape" and T5: "Virtual Reality";

- the mood M5: "Romance" whose weighting WMS is equal to 5.5 with the theme T6: "Forbidden love";

- the mood M6: "Tears" whose weighting WM6 is equal to 4.5 with the theme

T7: "Mourning and Bereavement";

- the mood M7: "Subtle" whose weighting WMZ is equal to 4 with the theme

T8: "Brain bend".

The user interacts with one or more contents. The content or contents with the user (s) who interacted are identified (step 120) by the recommendation system. The interaction is performed by means of a system interface, such as a touch screen or mouse or keyboard type device, which allows for example to click on the content.

In step 130, pairs of mood (s) and / or theme (s) are generated to form the semantic imprint of the user. Said pairs are generated from mood (s) and / or theme (s) of the contents with which the user has interacted.

The pairs of mood (s) and / or theme (s), the weights of said pairs, and possibly the associated weighting vectors, which form the semantic fingerprint of the user are stored in a user file or database.

The generation of a pair of the semantic imprint of the user from a mood or two moods present (s) in a content with which the user has interacted, is achieved only if the or each of said moods is a main mood. A main mood is a mood whose weighting value is strictly greater than a predefined value. In the example illustrated in the figures, said threshold value is 5.5. Thus when a pair of the semantic imprint of the user comprises one or two moods, the or each of said moods corresponds to a main mood. However, pairs are also generated from the themes of said content even though said themes are not associated with main moods.

Thus, as illustrated in FIG. 2, when the user views the content C1, the system detects the WTCH interaction and generates the pairs P1 M, P2M, P3M P4M,. .. P20ii, ... from the main moods and themes of the content C1.

In step 140, the weights of the pairs are determined as a function of the value associated with the interaction which concerns said pairs. In the example illustrated in FIG. 2, the value of the WTCH interaction is 3. Advantageously, a weighting increase is attributed to the weighting of a pair of the semantic imprint of the user whose mood is a main mood and / or whose theme is associated with a main mood.

Thus the weightings of P1 to P3 are each 3 ^* 1, 3 = 3.9 since they benefit from the weighting coefficient fixed here at 1, 3, whereas the weights of P4 and P20 are of 3 since their themes are associated to secondary moods.

From this footprint as shown in Figure 2, the contents can be ordered in step 150 according to the weighting pairs of the footprint. This scheduling is detailed below in the case of an additional interaction.

User interaction with content may be one or more of the following events:

reading of said content, for example for viewing and / or listening; and or

- adding to a reading list; and or

- alerting said content; and or

- the recording of said content; and or

- the notation of said content.

Each interaction is equated with a level of interest for the user. Each level of interest thus detected by the recommendation system is translated into a weighting, which determines or modifies the weighting of one or more pairs of the semantic imprint of the user.

As illustrated in Figure 3, the C2 content that corresponds to the movie "Spiderman" includes the following themes and moods: the M2 mood: "Spectacular" whose weighting is equal to 9.5;

- the mood M3: "Suspense" whose weighting is equal to 5.5 with the theme T3: "Beat the dock";

- the mood M4: "Strange" whose weighting is equal to 5, with the themes

T9: "Monsters" and T10: "Superheroes";

- the mood M5: "Romance" whose weighting is equal to 5 with the theme

T1 1: "Love in Action" and the theme T6: "Forbidden Love"

- mood M8: "Friends" whose weighting is equal to 3 with the theme

T12: "Teenager and College kids"

- the mood M7: "Subtle" whose weighting is equal to 0 with the theme T13:

"Scientists"

- the mood M9: "Places" whose weighting is equal to 0 with the theme

T14: "New York"

In addition, the content C2 also includes an additional parameter, called the weighting vector, denoted VTIO. The weighting vector targets a theme and / or a mood. In the example illustrated in the figures, the weighting vector VTIO targets the theme T10 so that the weighting vector will be associated with the pairs P200, P201, P215, called vectorized pairs, other than those comprising the targeted theme T10, which will be generated in the semantic imprint on the basis of C2 content.

Subsequently, the score of a content within a vectorized pair, which includes said vectorized pair but does not include the targeted theme, here the theme T10, is then impacted by said vector as detailed below in connection with Figure 5.

As illustrated in FIG. 4, when the user assigns a note to the content C2, the system identifies in step 121 the user's RTE interaction with the content C2, which impacts the user's semantic imprint .

The interaction of the user with the content C2 causes the generation (step 131) in the semantic imprint of the user, of one or more new pairs comprising a pair of theme (s) and / or mood (s) present in the content C2.

The pairs Ρ200,2, P201, 2 and P215,2 are thus generated because of the presence of themes T1 1, T9, T13, T14 and T3 in the content C2.

The weighting of the new pair P215 is determined in step 141 from the value of the RTE interaction, namely 12. Since the pair P215 does not present a main mood or theme associated with a main mood, the value of the interaction is not increased for the weighting calculation.

Since the topics T1 1, T9, T13 are associated with main moods, the weighting of each of the new pairs P200 and P201 is determined by the value of the interaction, namely 12, with a coefficient of increase, here 1, 3. Thus, the weighting of each of these P200 and P201 pairs is 15.6.

As a result of the interaction RTE with the content C2, the weights of the pairs P1, P2, respectively composed of the themes T5, T8 on the one hand, T4, T3 on the other hand, remain unchanged because the themes T5, T8, T4, do not belong to C2 content.

A new interaction may also result in the modification (step 132) of one or more pairs of the user's semantic imprint. According to a particular aspect, the modification of the weighting of at least a portion of the previously generated pairs is performed following the interaction of the user with the new content when the latter presents one or more pairs of themes and / or mood (s) found in the previous content.

In particular, it is possible for the calculation of the weighting of a pair following an interaction with a carrier content of said pair to be performed by adding the value of said interaction, possibly affected by a coefficient of increase, to the previous weighting value of said pair.

Thus, the P4 pair weighting is increased by the value associated with the interaction RTE, namely 12, because the themes T3 and T6 are also present in the content C2, so that the weighting of the pair P4 goes to 15. However, the pair P4 does not benefit from a coefficient of increase because none of its themes is associated with a main mood in the content C2.

In connection with FIGS. 2 and 4, it will be noted that the index I1 and subsequently the index i2 corresponds to the number of interactions. Thus, the pair P4i1 corresponds to the pair P4 defined following the first interaction WTCH, and the pair P4i2 corresponds to the pair P4 defined following the second interaction RTE.

Advantageously, when two pairs of the semantic imprint of the user have the same weight, the two pairs are ordered relative to each other following the alphabetical and / or numerical ascending order of their designation. Thus, in the example illustrated in FIGS. 4 and 5, the pair with the highest weight is considered to be the P200 pair whose themes are T9 and T1 1.

The contents C3 corresponding to "Jack", C4 corresponding to "Thor", and C5 corresponding to "Hulk" are associated with the pair P200 since they each contain the themes T9 and T1 1. The contents C3, C4 and C5 are ordered within the pair P200 according to their score. As mentioned above, the score of each of these contents can be a popularity value.

As illustrated in FIG. 5, the content C3 does not include the theme T10. Since the weighting vector VT10 is associated with the pair P200, it follows that the initial score of the content C3 is lowered by applying to said initial score a coefficient of between 0 and 1, here 0.1.

The C4 content whose score is 3.5 is thus recommended in front of the C5 content whose score is 3.2 and C3 content whose score has been lowered to 0.34.

The scheduling of the contents with regard to the different pairs of the semantic imprint of the user is carried out so as to recommend contents of a given weighting pair in front of contents of a lower weighting pair. In other words, the recommendation method gives priority to contents containing themes and / or moods corresponding to a pair of strong weightings present in the user's footprint in relation to the score of the contents which is a secondary criterion of scheduling.

Advantageously, when the contents of a given pair are recommended, the other pairs that present one of the theme (s) and / or mood (s) of said pair are excluded. Note that the contents themselves that correspond to said pairs are not excluded so that, if these contents correspond, by their theme (s) and / or mood (s), also to another pair not excluded, they can be recommended within such another pair not excluded.

In the example illustrated in the figures, the pair P201 and the pair P200 whose contents are recommended as explained above have in common the theme T1. The pair P201 is then excluded in step 142 of the recommendation process. The recommendation then continues on the basis of the remaining pair of highest weight, here the pair P4.

Thus, the recommendation of the contents C5, C4, C3 of the pair P200 (FIG. 6) is followed in step 151 of the recommendation of the contents C6 and C7 which comprise the themes T3, T6 and are thus associated with the pair P4. (Figure 7). The content C6 corresponds to the film "Charleston and Vendetta" and the content C7, of lower score, corresponding to the film "Hancock".

Thus the contents C1, C2, C3, C4, C5, C6, C7 are ordered by putting firstly the contents C1, C2, C3, C4, C5, presenting a pair of theme (s) and / or mood (s). corresponding to the pair P200 of the semantic fingerprint of the user of greater weighting, in front of other contents C6, C7 having a pair of theme (s) and / or mood (s) corresponding to a pair P4 of the semantic fingerprint of the user of lower weighting. And even within a given pair, the contents are ordered according to their score which can be impacted by a weighting vector associated with said pair.

The operations described above can be performed in the form of a computer program comprising program code instructions for the execution of said steps by a processor forming part of the computer system which can be for example a computer, a television Smart, a tablet or a smart phone. In particular, the various means of the system, which make it possible to obtain said functions of allocation, identification, generation, determination, scheduling, exclusion, increase, weighting can be carried out in the form of modules computer executable by a system processor, possibly on a remote server means of displaying the system. The system also includes a display screen and a communication interface with the user such as a keyboard and / or a mouse and / or a touch interface.

These computer programs, or computer instructions, may be contained in program storage devices, such as computer-readable digital data storage media, or executable programs. Programs or instructions can also be run from program storage devices.

In particular, the system for implementing the method may comprise electronic and / or computer parts configured to perform the operations described above. According to one embodiment, the system comprises a storage memory of a program comprising instructions for implementing the method, a processor for executing said program, a human-machine interface allowing the user to interact with content, a communication module with the content database and a communication module with the user file or database, an ordered content display screen for recommending said contents to the user. Of course the results of the operations presented above can be stored in one or more memories of the system In addition, it can be provided that all or part of the system is accessible via a network, for example via the Internet.

Although at least one embodiment of the invention has been illustrated and described, it should be noted that other modifications, substitutions and alternatives occur to those skilled in the art and may be changed without departing from the scope. of the object described here.

The present application contemplates covering all the adaptations and variations of the embodiments described above. Moreover, the term "including" does not exclude other elements or steps. In addition, features or steps that have been described with reference to one of the embodiments set forth above may also be used in combination with other features or steps of other embodiments set forth above. Note that it is necessary to include in the scope of the patent all the modifications envisaged above insofar as they form part of the contribution of the inventors to the prior art. Such modifications, substitutions and alternatives can be made without departing from the scope and spirit of the present invention.

Claims

1. Method for recommending content to a user, implemented using a computer system,

to each of the contents (C1, C2, C3, C4, C5, C6, C7) being allocated:

□ first parameters (T1, ..., T30) of themes, called themes, and

□ second mood parameters, called moods (M1, ..., M9), each theme being associated with one or more moods;

said method comprising the following steps:

- identifying (120) an interaction, called the first interaction (WTCH), of the user with one of the contents (C1), called first content;

generation (130), according to said first interaction (WTCH) and said first content (C1), of a set, called the user's semantic imprint, which comprises pairs (P1, P2, P3, P4,. .., P20) of mood (s) and / or theme (s) defined according to the themes and moods of said first content (C1);

- determining (140) for each of said pairs (P1, P2, P3, P4, ..., P20) of the semantic fingerprint of the user, a weight (WPI, WP2, WP3, _WP4 .. ., Wp2o) defined according to said first interaction (WTCH);

a step of identifying (121) a second interaction (RTE) of the user with a content (C2), called second content; and

□ a generation step (131) in the semantic fingerprint of the user, of at least one new pair (P200, P201, P215) corresponding to a theme pair (s) (T9, T1 1, T13, T14 ) and / or mood (s) present in the second content (C2), and a step of determining (141) the corresponding weighting (WP2OO, WP201, WP2IS) of the new pair as a function of said second interaction (RTE);

And / or, a modification step (132), as a function of the second interaction (RTE) with the second content (C2), of the weighting (Wp ₄ ) of at least one pair (P4) of said semantic fingerprint of the user,

- determining a value, called score, for each content (C1, C2, C3, C4, C5, C6, C1);

- scheduling of the contents (C1, C2, C3, C4, C5, C6, C7) by putting priority ahead the contents (C3, C4, C5) which present a pair of theme (s) and / or mood (s) corresponding to the pair of the semantic fingerprint of the user of greater weighting, compared to other contents (C6, C7) having a pair of theme (s) and / or mood (s) corresponding to a pair (P4) of the semantic fingerprint of the lower weighted user,

the contents (C3, C4, C5) associated with the same pair (P200) being ordered together according to their score,

2. Method according to claim 1, characterized in that

an additional parameter, called weighting vector (VTIO) being assigned to at least one (C2) of the contents with which the user interacted, said weighting vector targeting a given theme (T10) or mood, the method comprises the step of associating said weighting vector (VT10) with the pairs (P200, P201, P215) of the fingerprint, called vectorized pairs, other than those comprising the targeted theme (T10), which are generated in the semantic fingerprint based on said content (C2),

and in that the score of a content (C3) which comprises one of said vectorized pairs (P200), is determined according to the presence or absence in said content (C3) of the theme (T10) or of the target mood by the weighting vector

3. Method according to one of the preceding claims, characterized in that,

a weighting (WMI, ..., WM9) being assigned to each mood of a content according to the themes that are associated with this mood and attributed to said content, and

a mood (M1, M4) of a content (C1) which has a weighting value beyond a given threshold value, being called the main mood, the weighting determining step for each pair of the fingerprint includes assigning a markup to the weighting of a pair (P1, P2, P3) of the semantic imprint of the user, when one of the moods composing said pair is a main mood, or when the one of the themes composing the said pair is associated with a main mood.

4. Method according to one of the preceding claims, characterized in that,

a mood (M1, M4) of a content (C1) which has a weighting value beyond a given threshold value, being called the main mood, the generation of a pair (P3) of the semantic fingerprint of the user from a mood (M1) or two moods present (s) in a content (C1) with which the user has interacted, is performed only if the or each of said moods (M1) corresponds to a main mood.

5. Method according to one of the preceding claims, characterized in that said method comprises a pair recommendation exclusion step (142) according to which the recommendation of contents associated with a first pair (P200) of the semantic fingerprint of the user, causes the exclusion of a second pair (P201) of the semantic imprint of the user who presents a theme (T1 1) or a common mood with a theme or a mood of the first pair (P200) the semantic imprint of the user.

6. Method according to one of the preceding claims, characterized in that said method comprises a content exclusion step relative to a pair of the semantic fingerprint of the user, according to which the recommendation of content (s) associated ( s) to a first pair of the semantic imprint of the user leads to the exclusion of this (these) same content (s) in a second pair of the semantic imprint of the user to whom is or are also associated with the said content (s).

7. Method according to one of the preceding claims, characterized in that the weighting of a pair of the semantic fingerprint of the user, generated as a function of an interaction with a content, is a function of the type of the interaction. identified.

8. Method according to one of the preceding claims, characterized in that said method comprises storing filter data, for example temporal data, such as a consultation schedule and / or, the date and / or time implementing said method of recommendation, according to which the contents are ordered.

9. Method according to one of the preceding claims, characterized in that the interaction of the user with at least one (C1, C2) contents comprises:

reading of said content, for example for viewing and / or listening; and or

- adding to a reading list; and or

- alerting said content; and or

- the recording of said content; and or

- the notation of said content.

A computer program comprising program code instructions for performing the steps of a method according to one of claims 1 to 9 when said program is executed by a processor forming part of a computer system, such as a computer, a smart TV, a tablet or a smart phone.

1 1. Computer system for recommending to a content user, each content (C1, C2, C3, C4, C5, C6, C7) being allocated:

□ first parameters (T1, ..., T30), called themes, and

□ second parameters, called moods (M1, ..., M9),

each theme being associated with one or more moods

said system comprising:

means for identifying (120) an interaction, called the first interaction, with a content (C1), called the first content, with which the user has interacted; generation means (130), as a function of said first interaction and said first content (C1), of a set, called the user's semantic fingerprint, which comprises pairs (P1, P2, P3, P4,. .., P20) of mood (s) and / or theme (s) defined according to the themes and moods of said first content;

determining means (140) for each of said pairs of the imprint of a weighting defined as a function of said first interaction;

means for identifying (121) a second interaction (RTE) of the user with a content (C2), called second content; and

□ generation means (131) in the semantic fingerprint of the user, of at least one new pair (P200, P201, P215) corresponding to a theme pair (s) (T9, T1 1, T13, T14 ) and / or mood (s) present in the second content (C2), and a step of determining (141) the corresponding weighting (WP2OO, WP2OI, Wp2i5) of the new pair as a function of said second interaction (RTE);

And / or, modification means (132), as a function of the second interaction (RTE) with the second content (C2), of the weighting (Wp ₄ ) of at least one pair (P4) of said semantic fingerprint of the user,

means for determining a value, called score, for each content;

means for scheduling the contents (C1, C2, C3, C4, C5, C6, C7), with priority being given to the contents (C3, C4, C5) which present a pair of themes and / or moods (s) corresponding to the pair (P200) of the semantic fingerprint of the user of greater weighting, compared to other contents (C6, C7) presenting a pair of theme (s) and / or mood (s) ) corresponding to a pair (P4) of the semantic fingerprint of the lower weighting user,

said scheduling means being configured also to order between them the contents (C3, C4, C5) associated with the same pair (P200) according to their score

12. Computer system according to claim 1 1, characterized in that, an additional parameter, called weighting vector (VTIO) is assigned to at least one (C2) of the contents with which the user has interacted, said vector of weighting targeting a theme (T10) or a given mood,

the system comprises means for associating said weighting vector (VT10) with the pairs (P200, P201, P215) of the imprint, called vectorized pairs, other than those comprising the targeted theme (T10), which are generated in the semantic imprint on the basis of said content (C2), and in that, for a content (C3) which comprises one of said vectorized pairs (P200), said score determining means are configured to determine the score of said content (C3 ) depending on the presence or absence in said content (C3) of the theme (T10) and / or the target mood of the weighting vector (VTIO).

13. System according to one of claims 1 1 and 12, characterized in that,

a mood (M1, M4) of a content (C1) which has a weighting value beyond a given threshold value, being called the main mood, the weighting determination means for each pair of the fingerprint are configured to assign a weight increase to the weighting of a pair (P1, P2, P3) of the user's semantic fingerprint, when one of the moods composing said pair is a main mood, or when one of the themes component said pair is associated with a main mood.

14. System according to one of claims 1 1 to 13, characterized in that,

a mood (M1, M4) of a content (C1) that has a weighting value beyond a given threshold value, being called the main mood, the pair generation means are configured such that the generation of a pair (P3) of the user's semantic fingerprint from a mood (M1) or two moods present (s) in a content (C1) with which the user has interacted, is performed only if the or each of said moods (M1) corresponds to a main mood.