A method for processing digital content that conveys a narrative of an experience, and applications of this method for recommending content and / or experiences.

Abstract

"Method for processing digital content expressing a narrative of an experience, and applications of this method for content and / or experience recommendation." A computer-implemented method for processing digital content expressing a narrative of an experience lived and / or imagined by a user, comprising the following steps: capturing at least one piece of digital content expressing a narrative of an experience lived and / or imagined by said user; processing said digital experience narrative content by segmenting it into n narrative segments, each with a homogeneous tone; constructing a graph associated with said experience by concatenating graphical representations, each corresponding to one of said n narrative segments; comparing said experience graph with a set of pre-established elementary graphs, each corresponding to a specific archetype; and identifying the archetype corresponding to said experience. See Figure 1

Description

Title of the invention: Method for processing digital content expressing a story of an experience, and applications of this method for the recommendation of content and / or experiences FIELD OF THE INVENTION

[0001] The present invention relates to a method for processing digital content expressing a story of an experience lived and / or imagined by a user. It also aims at applications of the processing method for the recommendation of content or experience as well as for connecting several users. STATE OF THE ART

[0002] In all marketing strategies, there is a need for brands to segment their customers, in order to better communicate with them, adapt the product and experiential offer to their needs, recommend content to them, target them with advertising and / or refine artificial intelligence models to better predict commercial or business data.

[0003] Most traditional methods involve segmenting customers based on geographic, demographic, socio-cultural, behavioral, and / or customer habits and purchasing behaviors. Digital data such as website history, movie and TV viewing history, pages or articles liked, customer interests, and / or profile information - such as birthdays - can also help target audiences.

[0004] However, with the popularization of social networks and the omnipresent feedback culture, many customer comments, consumer opinions and stories are published every day on the Internet and / or shared directly with brands. Most of the time these comments are written by users eager to share their lived experiences. These comments can take the form of writings, writings composed of emoji or other figurative forms representing their emotions but also short videos and shared images. They are most often associated with a rating of the experience, for example in the form of stars. We start here from the principle that an experience shared by voice can be easily transcribed into text via STT (speech-to-text).

[0005] Since then, new customer segmentation methods based on comments have been proposed.

[0006] First of all, the most common method uses the rating given to the experiences: for example, we create a first cluster of people who gave 5 / 5, then a second cluster of people who gave 4 / 5, then 3 / 5, 2 / 5 and 1 / 5.

[0007] The second method uses the NPS (Net Promoter Score). Customers are asked to rate their experiences on a scale of 0 to 10 (or 1 to 10). People who give a score of 6 or less are called "Detractors", those who give 7 or 8 are called "Passives", while those who give 9 or 10 are "Promoters". The NPS is calculated by: % Promoters - % Detractors x 100. We therefore group people into 3 categories: Promoters, Passives and Detractors. Note that the Customer Effort Score (CES) is an indicator of customer experience inspired by the principle of the NPS and promoted within the Harvard Business Review in 2010 [1],

[0008] The CES follows a different logic from that of the NPS because it does not directly measure satisfaction, but an overall estimate of the effort provided by the customer.

[0009] The third method is to use sentiment analysis and this time focuses on content written by customers (in English: sentiment analysis). For example, people who have a rather positive opinion (value between 0 excluded and 1 of the sentiment analysis) will be grouped together. The same goes for those who had a rather negative sentiment (value between -1 and 0 excluded). Values ​​around 0 can be associated with people who have a neutral opinion. Note that many algorithms exist to assign a "sentiment score" ("for sentiment rating") to written comments and that they can also qualify a positive sentiment: for example, if a customer writes "delicious" the sentiment score will be around 0.8 (close to 1) while "correct" will be around 0.2.

[0010] However, it must be noted that most of the methods mentioned above are interested in the result of the experience (the rating given with stars, the NPS / CES) or their overall feeling (positive, negative, neutral) but not in the actual content and the story of these people. Consequently, the segmentation is not very precise and difficult to represent.

[0011] This system is based not on the overall rating given by customers but on what they have said.

[0012] Document US9762733B1 discloses a system for recommending a communication medium for interacting with a customer of a contact center, this system comprising: a processor; and a memory, wherein the memory contains instructions which, when executed by the processor, cause the processor to: detect a first interaction via a first medium with the customer; identify a plurality of communication mediums based on constraints to determine one or more candidate communication mediums that are different from the first medium for a second interaction with the customer; for each communication medium of the candidate communication mediums, estimate an expected value to be obtained using the communication medium for the second interaction; and select a particular communication medium from the candidate communication mediums based on the estimates to establish the second interaction via the selected communication medium.

[0013] Today, user recommendations are made using relational graphs. We recommend users who are closest to us socially, or who have friends in common, who are in the same groups, in the same circle of relationships around us, etc. Facebook, for example, uses "social graphs" to recommend users.

[0014] We also recommend users because they liked our profile and we liked it in return, or because we have common affinities, because they are the most desirable or the most "desirable" (Elo score), the most active, or the most popular (example Tinder).

[0015] Also because they correspond to our preferences, our interests, our social category, or simply because they are the closest geographically.

[0016] A combination of these criteria can also be used. However, it is clear that we cannot limit human beings and social relationships to this type of criteria.

[0017] Life being a series of experiences, each human being is shaped by the experiences he lives, by the stories he lives, regardless of the content of these experiences, as well as by the experiences he wishes to live in the near future.

[0018] The aim of the invention is to propose a new concept for processing digital content expressing a story of experience lived and / or imagined by a user, which overcomes the limitations and drawbacks of the processing methods of the prior art. Another objective of the invention is to propose applications for processing digital content expressing a story of experiences making it possible to propose tools for recommending content or for connecting users.

[0019] DEFINITIONS

[0020] Digital content: any type of digital file, whether audio, textual or visual

[0021] Account of an experience: sequential restitution in narrative form of an experience

[0022] NLP Natural Language Processing or TALN (Natural Language Automatic Processing)

[0023] Tone: the known technique known as sentiment analysis makes it possible to determine the tone of a text in order to see whether it evokes a positive, negative or neutral feeling.

[0024] Graph: a set of vertices connected by a set of arcs or edges

[0025] Archetype: Ideal type serving as a model. It can also represent a symbol in any form, which may appear in different cultures.

[0026] Sentiment Score: A sentiment score can be calculated by algorithms that evaluate the tone of a transcript on a spectrum from positive to negative. The score can, for example, be calculated as the ratio of positive to negative statements in the entire call, on a score of -1 to +1, with zero being neutral.

[0027] Emotion Dictionary: A compilation of words, graphical representations and icons (such as emojis) used to express feelings and emotions, for example in the form of a database. Statement of the invention

[0028] This objective is achieved with a computer-implemented method for processing digital content expressing a narrative of an experience lived or imagined by a user, comprising the following steps: a. capture at least one digital content expressing a story of an experience lived and / or imagined by said user, b. processing said digital content expressing said experience narrative, by segmenting it into n narrative segments each presenting a homogeneous tone, c. constructing a graph associated with said experience, by concatenation of graphic representations each corresponding to one of said n narrative segments, d. compare said experience graph with a set of previously established elementary graphs, each corresponding to a specific archetype, and identify the archetype corresponding to said experience.

[0029] Digital content here means any type of digital file, whether audio, textual or visual, expressing a narrative of experience.

[0030] In the method according to the invention, the experience narrative is processed and segmented into n narrative segments, each having a homogeneous tone. A graphical representation is then assigned to each type of tone, for example: - Positive tones will be associated with portions of the curve with a positive slope - Negative tones will be associated with portions of the curve with a negative slope

[0031] However, the present invention is not limited to this type of graphic representation:

[0032] We can also consider graphic representations using simple geometric shapes such as squares, circles and triangles, for example: - Positive tones will be associated with squares □ (by convention) - Negative tones will be associated with A triangles

[0033] A story of experience with a (+ / -) tone will therefore become: □ A, while a story with a (+ / - / -) tone will be transformed into the following sequence of forms: □ AA

[0034] These graphical representations may also use complex geometric shapes, such as 2D, 3D or 4D structures.

[0035] Finally, the use of icons, symbols, gifs and emoji is also possible.

[0036] The step of processing the digital content expressing a story preferably comprises a step for assigning a tone to each segment by an artificial intelligence previously trained on a corpus of labeled segments.

[0037] The step of processing the digital content expressing the story may advantageously comprise a step for detecting in said story content one or more breaks in tone, each break in tone initiating a new graphic representation different from that of the previous graphic representation.

[0038] The step of detecting tone breaks may preferably implement artificial intelligence.

[0039] The tones assigned to the story segments include, for example, a positive tone and a negative tone.

[0040] When the graphical representations use curve portions, the graph construction step may then further comprise a step for assigning a first predetermined slope to any graphical representation corresponding to a segment or a series of segments having a positive tone and a second predetermined slope to any curve portion corresponding to a segment or a series of segments having a negative tone.

[0041] The tones assigned to the narrative segments may further include a neutral tone.

[0042] The graph construction step may further comprise a step for assigning a substantially zero slope to any portion of the curve corresponding to a segment or a series of segments having a neutral tone.

[0043] The step of processing a digital content of a user's experience story can further be configured to determine in said story one or more adjuvants and one or more opponents according to an actantial schema.

[0044] In a first embodiment of the invention, one or more opponents and / or one or more adjuvants can be determined by artificial intelligence applied to the segments of the experience narrative.

[0045] In a second embodiment of the invention, one or more opponents and / or one or more adjuvants may be determined by a natural language processing (NLP) tool.

[0046] In a third embodiment of the invention, one or more opponents and / or one or more adjuvants are determined respectively by identification in the account of experiencing one or more negative emotions and / or one or more positive emotions.

[0047] The processing method according to the invention may further comprise a step for determining, for a user's experience report, a ratio of the number of adjuvants to the sum of the number of adjuvants and the number of opponents.

[0048] When the processing method according to the invention is applied to an experience for which the user has issued an experience report and an experience rating, it further comprises a step for graphically positioning said experience on a matrix presenting on the ordinate a measurement of the experience rating and on the abscissa a measurement of the ratio of the number of adjuvants to the sum of the number of adjuvants and the number of opponents.

[0049] When the processing method according to the invention is applied to a plurality of experiences for each of which the user has issued an experience report and an experience note, characterized in that said experiences are graphically positioned on the same matrix.

[0050] When the processing method according to the invention is applied to a plurality of users each having had several experiences, said experiences are graphically positioned on the same matrix.

[0051] According to another aspect of the invention, an application of the processing method according to the invention is proposed, for the recommendation of content to a user, comprising the following steps: a. capture at least one digital content expressing a story of an experience lived and / or imagined by said user, b. processing said experience narrative content, by segmenting it into n narrative segments each presenting a homogeneous tone, c. construct a graph associated with said experience, by concatenation of graphic representations each corresponding to one of said n story segments, d. compare said experience graph with a set of previously established elementary graphs, each corresponding to a specific archetype, and identify the archetype corresponding to said experience, e. extract from one or more databases, at least one content corresponding to the said archetype thus identified, f. transmit to said user information recommending said at least one content thus extracted.

[0052] The processing method according to the invention may further comprise a step for processing a plurality of experience graphs already constructed for the same user, said processing step comprising: - for each elementary graph, a determination of the number of said experience graphs which are identical to said elementary graph, - a determination of the respective ratios of elementary graphs in said plurality of experience graphs, - a classification of said user in a group of users presenting substantially the same elementary graph ratios,

[0053] the recommendation step being configured to recommend the same content to users belonging to said group of users.

[0054] For an application in which the recommended content comprises recommendation content of an experience lived and / or imagined by a first user, the recommendation step is configured to transmit said experience recommendation content to at least one other user sharing at least one archetype with said first user.

[0055] According to yet another aspect of the invention, an application of the processing method according to the invention is proposed, for recommending a connection between users, comprising the following steps: - capture a plurality of digital contents expressing stories of experiences emitted by a first user previously invited to live and / or imagine a plurality of predefined experiences from a set of experiences, - process each of said experience narratives, segmenting each into n narrative segments, each presenting a homogeneous tone, - construct for each of said experience narratives a graph associated with said experience, by concatenation of graphic representations each corresponding to one of said n story segments, - compare each of said experience graphs with a set of previously established elementary graphs, each corresponding to a specific archetype, so as to identify for each experience narrative the archetype corresponding to said experience, - reproducing said aforementioned steps of capture, processing, construction and comparison for a predetermined set of other users, so as to attribute to each of said other users a plurality of archetypes corresponding to their experience narratives, - determining for each other user a similarity rate with said first user, as being the ratio of the number of archetypes common with said first user to the total number of archetypes identified from the plurality of experiences, - transmit to said first user information recommending connection with other users having a similarity rate greater than a predetermined similarity threshold.

[0056] The present invention is based on the fact that people who have experienced similar stories in the past and / or who have the same aspirations for the future are more likely to like each other, share common interests and form emotional bonds. The way in which these people will tell their stories is as important as the content and outcome of their experiences, and allows people to be grouped together within common archetypes.

[0057] This involves quickly and reliably proposing relevant users to each of these individuals. This process can be useful for dating sites and recommending profiles of people who have a great experiential similarity to us. Or also within social networks to propose new "friends" (Facebook, LinkedIn), or community platforms to recommend new users.

[0058] In an advantageous embodiment, the graph of each of the experiences in step e) is constructed by segmenting the customer story into n segments, a tone (positive, negative, neutral; neutral being optional) then being assigned to each of these segments by an Artificial Intelligence (AI), previously trained on a corpus of labeled segments, then by identifying the tone breaks ((+ to -), (- to +), (+ to n to -), (- to n to +)) and by assigning a portion of curve with a positive straight line slope when the tone of a segment, or of a series of segments, is positive, and a portion of curve with a negative straight line slope when the tone of a segment, or of a series of segments, is negative. We will call these graphs “experience graphs”.

[0059] The calculation of the similarity rate of the profile of each user compared to the reference user in step f) can also take into account the psychological and behavioral similarity and compatibility of the users, by comparing their psychological profiles and their personalities.

[0060] The term "user" in step a) may refer to a user composed of two or more people interacting with the same electronic object (such as, but not limited to: tablet, phone, watch, voice speaker, device in a car).

[0061] This application further comprises a step for selecting all of the other users from a database of other users, based on criteria of geographical proximity to the first user.

[0062] The first user and / or the other users may comprise machines and / or robots and / or other intelligent systems such as artificial intelligences (AI).

[0063] In particular, it is advantageous to provide an artificial intelligence with “unsupervised learning” capable of understanding and determining the tone of each segment automatically without or with little prior training.

[0064] The step of transmitting connection recommendation information may comprise a vocal communication of this recommendation information to the first user.

[0065] The step of determining the similarity rate may include a weighting of the different experiences lived and / or imagined by the first user and / or by the other users.

[0066] Experience stories captured from other users are entered by voice recognition.

[0067] Experience stories captured from other users can be extracted from a blockchain and decrypted before being processed.

[0068] It is also possible to provide a step for transmitting to the first user information recommending connection with other users having a similarity rate lower than a predetermined maximum rate.

[0069] The step of capturing a plurality of experience narratives can furthermore be carried out on a set of first users interacting with the same connected electronic object.

[0070] The processing method according to the invention and its application for content recommendation use for this purpose a new customer segmentation which is based on elementary graphs of experiences. The postulate underlying the present invention is that people who have experienced similar stories are more likely to behave in the same way and therefore to belong to the same group of customers. From then on, similar content can be recommended to them.

[0071] By “content” we mean any type of communication (email, call, SMS, voice message, notification, etc.) but also any type of digital content (audio, music, podcast, video, film, series, image, text, book, product, experience, advertising, event, place, etc.).

[0072] In a preferred embodiment, the experience graph in step c) is constructed by segmenting the customer story into n segments, a tone (positive, negative, neutral; neutral being optional) then being assigned to each of these segments by an Artificial Intelligence (AI), previously trained on a corpus of labeled segments, then by identifying the tone breaks ((+ to -), (- to +), (+ to n to -), (- to n to +)) and by assigning a portion of curve with a positive straight line direction coefficient when the tone of a segment, or a sequence of segments is positive, and a portion of curve with a negative straight line direction coefficient when the tone of a segment, or a sequence of segments is negative.

[0073] In a particular embodiment of the processing method according to the invention, the length and / or the slope of each portion of the curve is varied as a function of the associated tones of the n narrative segments. The axes can be normalized between 0 and 1 or not.

[0074] The processing method applied to the content recommendation according to the invention may further comprise a step for analyzing a history of the experiences of each user in order to deduce a typical user profile and therefore to deduce a suitable content communication strategy.

[0075] The history of the experiments can be placed in the matrix below (experiment score vs. ratio of adjuvants / adjuvants + opponents) in order to deduce an adapted content strategy.

[0076] The processing method according to the invention applied to content recommendation may further comprise a step for analyzing experiences imagined and recounted in the form of writings or transcribed in writing (“In my ideal vacation, I would like to go camping, swim peacefully in the river but I would avoid being attacked by a bear”). Here we would have + / - tones, hence an “Icarus” type profile.

[0077] The experience history may further take into account lived and imagined experiences rather than only lived or only imagined experiences.

[0078] Segmentation by experiential archetypes of one of the previous methods (real time, real history, real history in matrix, imagined experience(s)) can be combined with at least one of the classic customer segmentation methods (experience rating, NPS / CES, sentiment analysis, use of geographic, demographic, socio-cultural, behavioral data, customer habits and purchasing acts, or digital data such as the history of websites consulted, the history of films and series viewed, pages or articles liked, customer interests, and / or information on their profile - such as birthdays). DESCRIPTION OF FIGURES

[0079] - [Fig.l] illustrates the main steps of the treatment process according to the invention, applied to the recommendation of contents; - [Fig.2] illustrates examples of archetype attribution for experiences lived by two users; - [Fig.3] represents an example of an experiment matrix implemented in the treatment method according to the invention; - [Fig.4] illustrates an example of an experience graph analyzed according to a Greimas actantial diagram; - [Fig.5] illustrates a first example of a break in tone between a segment of narrative with a positive tone and a segment of narrative with a negative tone; - [Fig.6] illustrates a second example of a break in tone between a negative-toned narrative segment and a positive-toned narrative segment; - [Fig.7] illustrates a set of experience graphs and archetypes implemented in an example of implementation of the treatment method according to the invention.; - [Fig.8] illustrates a particular example of experience graphs in which the curve portions have variable lengths; - [Fig.9] illustrates a set of experience narratives consisting of experience narratives all presenting the same sequence of tones (+ / -); - [Fig. 10] illustrates a set of experience narratives consisting of experience narratives presenting a series of tones (- / + / -) of different intensities. DETAILED DESCRIPTION

[0080] With reference to [Fig. 1], the recommendation method according to the invention uses a new customer segmentation based on elementary experience graphs. Our assumption is that people who have experienced similar stories are more likely to behave in the same way and therefore to belong to the same customer group. From then on, we can recommend similar content to them.

[0081] By "content" we mean any type of communication (email, call, SMS, voice message, notification, etc.) but also any type of digital content (audio, music, podcast, video, film, series, image, text, book, product, experience, advertising, event, place, etc.). We apply the process each time the customer shares their experience in real time.

[0082] The result of the experiential archetype calculated by the algorithm will trigger a targeted and personalized content recommendation operation for each customer. Client A: Claude Client B: Christine Experiment 4 Mowgli Experiment 9 Icarus

[0083] For example, in the case of Claude, if experience 4 has just been shared, we can consider that it was positive (positive tone, Mowgli archetype), The subsequent communication can be a customer loyalty operation via an email invitation to events or discount vouchers for their next order. This is more of a CRM (Customer Relationship Management) content.

[0084] In Christine's case, the system has just calculated that her experience 9 was mixed: although it started on a positive note, it ended on a negative note, hence an Icarus archetype (+ / -). Let's imagine that the cause is that there were a lot of people in the restaurant where she had this experience. Having analyzed the cause of this customer inconvenience can trigger an automatic email to Christine to solve her problem (for example, invite her at a time when there will be fewer people) or simply send her an email apologizing for her inconvenience with the automatic refund of her experience, in order to guarantee the best customer satisfaction in real time.

[0085] From then on, it appears possible for brands to target experiential archetypes in particular. For example, all the people who experienced negative things at the beginning but ended on a positive note, to target the "pain points" (problems) to be addressed, target CRM accordingly, look at which profiles are associated with such purchasing acts or build a particular strategy (we build loyalty with Robin Hoods, we call Romeo and Juliets by phone, and we offer discounts to Icarus).

[0086] In short, the method for processing experience narratives according to the invention consists of analyzing the experiences lived by a target customer base in real time, in order to determine a content recommendation strategy adapted to the experience they have lived.

[0087] The content recommendation method according to the invention may further comprise a step for analyzing a history of the experiences of each customer in order to deduce a typical customer profile and therefore to deduce a suitable content communication strategy, as illustrated by the example in [Fig.2].

[0088] From the archetype data represented in [Fig.2], it can be deduced that: - Claude is 75% Mowgli and 25% Icarus. - Christine is 67% Mowgli, 22% Icarus and 11% Robin Hood

[0089] Each of them will belong to a different client cluster for whom the content recommendation engine will push different content.

[0090] Analyzing this history of experiential archetypes can trigger a content recommendation operation directly for the customer. Let's imagine that Claude and Christine's experiences 3 and 4 were lived in similar places for the same type of experience. Since Claude and Christine liked these experiences very much, and Christine liked experience 5 very much, we could recommend Christine's experience 5 to Claude. Our process then becomes a content recommendation engine. Recommendation of inter-user experiences using the associated experience graph as a basis for comparison. It can also recommend experiential locations, that is, places in which we can have a specific experience, physical, digital, virtual or a combination of the previous types of experiences. Construction of an experience matrix

[0091] The history of the experiments can be placed in a matrix as represented in [Fig.3] (experiment score vs. ratio of adjuvants / adjuvants + opponents) in order to deduce an adapted content strategy.

[0092] A first method of calculating the ratio of adjuvants / adjuvants + opponents, consists of using the results of artificial intelligence affixing labels to the sentence segments (quest, adjuvants, opponents, glorifying stage, neutral), or the 3 labels (positive tone, negative tone, neutral tone). In our example, we have 3 “adjuvants” labels and 1 “opponents” label, the ratio is therefore 3 / 4.

[0093] The second method consists of analyzing the structure of the sentence, identifying common nouns, pronouns and proper nouns, and classifying them as positive, neutral or negative elements using NLP (Natural Language Processing). Thus the system will find as adjuvants: "the welcome" (warm), "the dishes" (very good), "value for money" (very good) and as opponents: the "restaurant" (small and not very flashy) hence the ratio adjuvants / adjuvants + opponents of 3 / 4.

[0094] The third method also analyzes the structure of the sentence but this time is only interested in the emotions and qualifiers used. For example, the algorithm will identify elements considered here as adjuvants: "warm", "very good", "fresh", "tasty", "spicy but not too much", "very good"; or opponents "small", "not very flashy". The ratio adjuvants / adjuvants + opponents will therefore be 6 / 8 = 3 / 4 = 0.75. We consider here that a positive emotion will be qualified as adjuvants, and that a negative emotion will be qualified as opponents.

[0095] In the case of Christine's nine experiments, we deduce the data shown on the matrix illustrated in [Fig.2].

[0096] Once each customer's experiences are positioned on this matrix, our system uses the clusters in the matrix to recommend content, such as CRM, or for the customer directly.

[0097] For example, if another person has their experiences positioned in the same areas of the matrix as Christine's, with the same associated archetypes, then they will be more likely to behave in the same way as Christine. We can therefore recommend the same type of content to them. Structure and composition of an experiment

[0098] Every comment posted on the Internet is first and foremost a story. A story that may be complete or incomplete, detailed or not, but a story. This has been demonstrated in literature by the thesis of Philippe Taupin [2] in which the author establishes that each lived or imagined experience has the structure of a story and therefore obeys the actantial schema of Greimas [3]. In the latter, a character, the hero, pursues the quest for an object. The positive characters, events, or objects that help him in his quest are called "adjuvants". The negative characters, events, or objects that seek to prevent his quest are called "opponents".

[0099] By transposing this narratology model to customer experiences, we consider a user customer (the hero) who pursues a quest (for example, discovering Italian cuisine - this is the reason which pushed the customer to live this experience), helped by adjuvants, and thwarted by opponents.

[0100] In addition to this: - Positive, neutral and / or negative emotions, - Senses mobilized (sight, hearing, touch, smell, taste - and all the variations - "heard", "felt", "music", etc.), - The 5 elements (wood, fire, earth, metal and water and all possible variations - "gold", "silver", "board", "rain", "drop", "puddle" ...) - Linguistic isotopies, that is to say according to Greimas [4], the redundancy of elements in a text allowing the latter to be understood, such as the redundancy of words in the same lexical field (e.g. rock, music, etc.). - And finally the result of the experience, which can be characterized by the grade given to the experience, thus what Greimas calls "the glorifying stage". Interpreting an Experience Graph

[0101] Each lived or imaginary experience therefore follows the model represented in [Fig.4] which represents the story of this experience in the form of a curve with ups and downs to signify that each story is not linear and therefore identical.

[0102] In light of the previously established model, it appears that customer segmentation based on the outcome of the experience is not precise because there may be different paths that lead to the same result. For example, a person who gives 4 stars may have experienced only positive elements, or on the contrary may have experienced many negative elements at the beginning but which were followed by a very positive element at the end of the experience, hence the rating. In the same way, an average sentiment score is of little interest because it will smooth out the experience by taking the average of what the customer experienced, without taking into account the highs and lows.

[0103] Drawing inspiration from the previous model and the work carried out by Kurt Vonnegut on the “shapes of stories” [5], as well as on the polarization of film and series scenarios According to McKee [6], we therefore propose to attribute graphs to comments and accounts of experiences, graphs that we will call “shapes of experiences”.

[0104] In the paper [7] “The emotional arcs of stories are dominated by six basic shapes” (Reagan, 2016), the authors analyzed over 1300 stories from Project Gutenberg literary works to derive 6 basic emotional arcs. The authors then examined whether there might be a correlation between story form and the success of the work (as measured by the number of downloads).

[0105] In the paper [8] “The Emotional Arcs of Television: NULab Project” (Chwe, 2019), the authors reused the techniques described in Reagan et al., 2016 and applied them to television. Since television is primarily a visual medium, they relied on subtitles to examine television programs and TV series through dialogue. The results were mixed.

[0106] The treatment method according to the invention proposes a radically new approach: - by applying the principle of emotional arc to everyday life experiences - because we know thanks to P. Taupin's thesis that an experience has the form of a story, - by segmenting said stories into sentence segments, to deduce a tone per segment and therefore an associated graph, and - by coupling these graphs together, in order to create 3D structures, feed a content recommendation engine, and ultimately offer content to users. Building an Experience Graph

[0107] a. Method 1: the segment and 1TA# method

[0108] The first method for constructing the graph of an experiment consists of several steps: - Segmentation of the customer story into n segments. - Categorization of segments by Artificial Intelligence (AI) according to 3 labels: positive tone, negative tone, neutral tone, the last one being optional. This categorization is done using an AI previously trained on a corpus of labeled segments. - Identification of tone breaks (+ to -), (- to +), (+ to n to -), (- to n to +)• - Construction of the graph of the analyzed experience. To do this, we take the tonality of the segments and the breaks in tonality. Each time the tonality of a segment, or of a series of segments is positive, the corresponding portion of the curve will have a positive slope of the straight line. Conversely, if the tonality of a segment, or of a series of segments is negative, the corresponding portion of the curve will have a negative slope coefficient.

[0109] Example: “The sky was beautiful but it was cold” is broken down as follows in Table I below: Segment No. Artificial Intelligence Prediction Result 1 The sky was beautiful Positive tone 2 But it was cold Negative tone

[0110] Table I

[0111] In this example, we obtain a graph as illustrated in [Fig.5]. a. Method 2: The “Nearest Neighbor Method”#

[0112] The second method for constructing the graph of an experiment consists of: - Directly analyze the sentence structure of each experience - Identify common nouns, pronouns and proper nouns - Give them a positive, neutral or negative tone depending on the context in which they are used - through the use of NLP (Natural Language Processing) and artificial intelligence (machine learning, deep learning) - Identify the tone breaks (+ to -), (- to +), (+ to n to -), (- to n to +), - Construct the associated experience graph.

[0113] Example: “The sky was beautiful but it was cold.” - “Heaven” and “he” are identified by the system. - Then, if we look at the words surrounding "ciel", we see "Le", "était" and "beau", which allows us to determine that "ciel" is associated with a positive tone (because of "beau"). - In the same way, the word "il" will be associated with a negative tone (because of "cold"), hence the same graph as previously. a. Method 3: the method of segments and the actantial model#

[0114] This third method is the most complete. It is based on the complete model established previously. 1. First, the customer story is segmented into n segments. The segmentation follows the grammatical structure of the sentence and / or punctuation, and is done automatically, with reference to Table II below. A very small restaurant, at least at first glance, and not very flashy either. But the welcome is warm and the Thai dishes (shrimp and squid salad, squid in green curry with coconut milk and rice) are very good, fresh, tasty and "spicy but not too much" Very good value for money I recommend

[0115] Table II

[0116] The size of the segments can vary and be limited to a word or a group of words. 1. Then, for each segment, we calculate / obtain, with reference to Table III below: - The segment's sentiment score. This sentiment score is obtained using classic functions from data science libraries. - The presence or absence of positive, neutral and / or negative emotions is determined by comparing the words in the segment with a dictionary of emotions classified into categories (+ / -, neutral). Surprise, for example, is considered a neutral emotion. Note that our dictionary of emotions also contains a database of emojis associated with positive, neutral and / or negative emotions. - Categorization of segments by Artificial Intelligence according to 5 labels: quest, adjuvants, opponents, glorifying stage, neutral. This categorization is done thanks to an Artificial Intelligence previously trained on a corpus of labeled segments. - We then calculate the tone of each segment. The goal is to determine whether the segment will be positive (+), neutral (n), or negative (-). A positive segment corresponds to a positive customer experience. And vice versa for a negative segment. N° Segment Sentiment score Dictionary of emotions Artificial intelligence Tone 1 A very small restaurant at least at first glance and not very flashy either 0.193 Opponents - 2 But the welcome is warm 0.0066 "warm" Adjuvants + 3 and the Thai dishes (salad with veal and calamari, calamari with green curry, coconut milk and rice) are very good, fresh, tasty and "spicy but not too much" 0.368 Adjuvants + 4 Very good value for money 0.7 Adjuvants + 5 I recommend 0.2 Glorifying step +

[0117] Table III

[0118] To calculate the tone, we compare the sentiment score, the emotions found in the emotion dictionary, as well as the labels applied by artificial intelligence. - If 3 elements are of the same sign (+ / -) or neutral, the tone is that of the 3 elements. In the case of segment #2, the sentiment score is 0.0066 (positive), “warm” is an emotion classified as positive, and “adjuvants” is considered positive by definition. Thus the tone of segment 2 will be positive. - If 2 elements are of the same sign (+ / -) or neutral, and the 3rd is different, the tone will be that of the two elements. - If the 3 elements are different, the tone of the segment will be that of the emotions. - If 2 elements are of the same sign (+ / -) or neutral, the tone will be that of the two elements. In the case of segments n°3, 4, the tone is therefore positive. - If 2 elements are different and the artificial intelligence returned a polarized result (adjuvants, opponents, neutral), the tone will be that of the artificial intelligence. In the case of segment n°1, the tone will therefore be negative. - If 2 elements are different and the artificial intelligence returned a non-polarized result (quest or glorifying step), the tone will be that of the sentiment score. In the case of segment #5, the tone is therefore positive.

[0119] The treatment method according to the invention is not limited solely to the previous case and can also operate: - Using only the sentiment score of each segment: the tone of the segment will then in all cases be that of the sentiment score of the segment. In this case, we would not use emotions or the results of artificial intelligence. - Using only the positive / neutral or negative emotions of the segments. The tone of each segment will be determined solely from the emotions. - Using only the results of artificial intelligence (AI) assigning the labels: quest, adjuvants, opponents, glorifying stage, neutral, without being limited to these labels. In this case, the tone of each segment will correspond to the results of the AI ​​per segment. - Using a combination of 2 of the 3 previous elements. For example, sentiment score and emotions; or sentiment score and AI results. 1. We then automatically analyze the tone breaks.

[0120] In other words, each time we pass: - from a + tone to a - tone - from a tone - to a tone + - from a + tone to an n tone then a - tone - from a tone - to a tone n then a tone +

[0121] In this example, there is only a break in tone between segment 1 (-) and segment 2 (+). 1. We then construct the graph of the analyzed experience. To do this, we take the tonality of the segments and the breaks in tonality. Each time the tonality of a segment, or of a series of segments is positive, the corresponding portion of the curve will have a positive slope of the straight line. Conversely, if the tonality of a segment, or of a series of segments is negative, the corresponding portion of the curve will have a negative slope of the straight line.

[0122] We then obtain an experience graph as illustrated in [Fig.6].

[0123] In a variant of the invention, the length and / or slope of each portion of the curve is varied according to the associated tones of the n narrative segments. The axes can be normalized between 0 and 1 or not.

[0124] For example, an experience narrative with tone (+ / + / + / -) or an experience narrative with tone (- / - / +) may have the forms as represented in [Fig.8]. Comparison of experience graphs

[0125] In the following description, we limit ourselves to three breaks in tone. We can construct, with reference to [Fig.7], a set of elementary graphs. The " / " identify the breaks in tone. These eight elementary graphs are associated with experiential archetypes: Mowgli, Romeo & Juliet, etc.

[0126] It should be noted that the names of these archetypes were attributed by referring to the stories of these heroes.

[0127] In our previous example, the archetype associated with a negative then positive slope is “Robin Hood”. Thus the customer who recounted this experience is part of the “Robin Hood” cluster.

[0128] In the case of this experience, the overall rating given by the customer is 4 / 5. We therefore demonstrate that the path to reach this satisfaction of 4 / 5 is not unique. Some will go through a Robin Hood type profile (- / +), others will be Mowgli (everything will be positive in their experience: +) while others may be also Cinderella (+ / - / +) or Peter Pan (- / + / - / +). To these eight elementary graphs, we add the “Neutral” archetype (neutral tone n - example “I am walking”).

[0129] In the context of the processing method according to the invention, it is possible to combine several experience narratives in order to form a “batch of experience narratives”, relating for example - but not limited to - the same experiential place, the same individual or the same line of products, or the same line of experiences.

[0130] We can then obtain 2D or 3D surfaces linked to this batch of experience stories. We can designate this new graphic object under the term "experience pattern". This "experience pattern" can have reliefs. The evolution over time of this indicator can help decision-makers to improve and optimize the customer experience in question. With reference to the example in [Fig.9], the batch of experience stories is relatively homogeneous because it is made up of experience stories all presenting the same sequence of tones (+ / -).

[0131] With reference to the example in [Fig. 10], the batch of experience reports is not homogeneous because it is made up of experience reports presenting a series of tones (- / + / -) of different intensities.

[0132] Of course, the present invention is not limited to the examples which have just been described. Many other examples of implementation of the processing method according to the invention and its applications for the recommendation of contents or for connections can be envisaged without departing from the scope of the present invention. REFERENCES

[0133] 1. Dixon, M., Freeman, K., & Toman, N. (2010). Stop trying to delight your customers. Harvard Business Review, 88(7 / 8), 116-122. 2. Taupin, P. (2017). Imaginaries of automotive atmosphere and semiotics of narratives: a digital experiential innovation approach in China (Doctoral dissertation, Pierre and Marie Curie University-Paris VI). 3. Algirdas Julien Greimas, Structural Semantics, Research and Method, Edition Larousse, 1996 4. AJ Greimas, On Meaning — Semiotic Essays, Seuil editions, 1970. Cited by A. Hénault, The Issues of Semiotics, PUF, 1993, p. 91. 5. Vonnegut, K. (1995). Shapes of stories. Vonnegufs Shapes of Stories, 51. 6. McKee, R. (1997). Story: style, structure, substance, and the principles of screenwriting. Harper Collins. 7. Reagan (2016). The emotional arcs of stories are dominated by six basic shapes. 8. Chwe (2019). The Emotional Arcs of Télévision: NULab Project.

Claims

Claims

1. A computer-implemented method for processing digital content expressing a narrative of an experience lived or imagined by a user, comprising the following steps: a. capturing at least one digital audio content expressing a narrative of an experience lived and / or imagined by said user, and segmenting this digital audio content thus captured into n segments, following the grammatical structure and / or punctuation of each sentence constituting the experience narrative, b. for each segment: - calculate a sentiment score, from data science libraries, - compare the words of said segment with a dictionary of emotions and assign said segment a positive, negative or neutral emotion, - categorize said segment by an artificial intelligence previously trained on a corpus of labeled segments, according to a predetermined number of labels associated with an actantial schema, c. calculate a tone of said segment, from the sentiment score, the emotion found in the emotion dictionary, and the actantial schema label, d. associate with each segment a predetermined geometric shape representative of the tone of said segment thus calculated, e. detect in said digital content one or more tone breaks, using artificial intelligence, a. constructing a graph associated with said experience, by concatenating the geometric shapes each corresponding to one of said n narrative segments, said experience graph thus constructed integrating said previously detected tone break(s), b. compare said experience graph thus constructed with a set of previously established elementary graphs, each corresponding to a specific archetype, and identify the archetype corresponding to said experience.

2. Method according to the preceding claim, in which the geometric shapes comprise straight line portions, characterized in that the graph construction step further comprises a step for assigning a first predetermined slope to any straight line portion corresponding to a segment or a series of segments having a positive tone and a second predetermined slope to any straight line portion corresponding to a segment or a series of segments having a negative tone.

3. Method according to the preceding claim, characterized in that the graph construction step further comprises a step for assigning a substantially zero slope to any portion of a straight line corresponding to a segment or a series of segments having a neutral tone.

4. Method according to any one of the two preceding claims, characterized in that each straight portion has a length and / or a slope which varies depending on the associated tones of the n narrative segments.

5. Method according to any one of the preceding claims, characterized in that the geometric shapes use graphic objects among icons, gif symbols and emojis.

6. Method according to any one of the preceding claims, characterized in that among the labels of the actantial diagram, one or more opponents and / or one or more adjuvants are determined by a natural language processing tool (“Natural Language Processing”) or NLP.

7. Method according to any one of the preceding claims, characterized in that among the labels of the actantial diagram, one or more opponents and / or one or more adjuvants are determined respectively by identification in the experience narrative of one or more negative emotions and / or one or more positive emotions.

8. Method according to any one of the two preceding claims, characterized in that it further comprises a step for determining, for a user experience narrative, a ratio of the number of adjuvants to the sum of the number of adjuvants and the number of opponents.

9. Method according to the preceding claim, applied to an experience for which the user has issued an experience report and an experience rating, characterized in that it further comprises a step for graphically positioning said experience on a matrix presenting on the ordinate a measurement of experience rating and on the abscissa a measurement of the ratio of the number of adjuvants to the sum of the number of adjuvants and the number of opponents.

10. Method according to the preceding claim, applied to a plurality of experiences for each of which the user has issued an experience narrative and an experience rating, characterized in that said experiences are graphically positioned on the same matrix.

11. Method according to the preceding claim, applied to a plurality of users each having had several experiences, characterized in that said experiences are positioned graphically on the same matrix.

12. Application of the processing method according to any one of the preceding claims, for the recommendation of digital content to a user, comprising the following steps: a. capturing at least one audio digital content expressing a story of an experience lived and / or imagined by said user, and segmenting this audio digital content into n segments, following the grammatical structure and / or punctuation of each sentence constituting the story of experience, b. for each segment: - calculating a sentiment score, from data science libraries, - comparing the words of said segment with a dictionary of emotions and assigning said segment a positive, negative or neutral emotion, - categorizing said segment by an artificial intelligence previously trained on a corpus of labeled segments, according to five labels

13. associated with an actantial schema: quest, adjuvant, opponent, glorifying stage, neutral. c. calculate a tone of said segment, from the sentiment score, the emotion found in the emotion dictionary, and the actantial schema label, d. associate with each segment a predetermined geometric shape representative of the tone of said segment thus calculated, e. detect in said digital content one or more tone breaks, using artificial intelligence, f. constructing a graph associated with said experience, by concatenating the geometric shapes each corresponding to one of said n narrative segments, said experience graph thus constructed integrating said previously detected tonal break(s), g. compare said experience graph thus constructed to a set of previously established elementary graphs each corresponding to a determined archetype, and identify the archetype corresponding to said experience, h. extract from one or more databases, at least one content corresponding to said archetype thus identified, i. transmit to said user information recommending said at least one content thus extracted. Application for content recommendation according to the preceding claim, characterized in that the processing method according to any one of claims 1 to 11 further comprises a step for processing a plurality of experience graphs already constructed for the same user, said processing step comprising: - for each elementary graph, a determination of the number of said experience graphs which are identical to said elementary graph, - a determination of the respective ratios of elementary graphs in said plurality of experience graphs, - a classification of said user in a group of users presenting substantially the same elementary graph ratios, the recommendation step being configured to recommend the same audio content to users belonging to said user group.

14. Application according to the preceding claim, in which the recommended audio content comprises audio content recommending an experience lived and / or imagined by a first user, characterized in that the recommendation step is configured to transmit said experience recommendation audio content to at least one other user sharing at least one archetype with said first user.

15. Application of the processing method according to any one of claims 1 to 11, to recommend a connection between users, a first user having issued a plurality of digital audio contents expressing stories of experiences, comprising the following steps: - for each of the said digital audio contents: a. segment this digital content into n segments, following the grammatical structure and / or punctuation of each sentence constituting the experience narrative, b. for each segment: - calculate a sentiment score, from data science libraries, - compare the words of said segment with a dictionary of emotions and assign said segment a positive, negative or neutral emotion, - categorize said segment by an artificial intelligence previously trained on a corpus of labeled segments, according to five labels associated with an actantial schema: quest, adjuvant, opponent, glorifying stage, neutral, c. calculate a tone of said segment, from the sentiment score, the emotion found in the emotion dictionary, and the actantial schema label, d. associate with each segment a predetermined geometric shape representative of the tone of said segment thus calculated, e. detect in said digital content one or more tone breaks, using artificial intelligence, f. constructing a graph associated with said experience, by concatenating the geometric shapes each corresponding to one of said n narrative segments, said experience graph thus constructed integrating said previously detected tonal break(s), g. compare said experience graph thus constructed with a set of previously established elementary graphs, each corresponding to a specific archetype, and identify the archetype corresponding to said experience, - reproducing said aforementioned steps for a predetermined set of other users, so as to assign to each of said other users a plurality of archetypes corresponding to their experience narratives, - determining for each other user a similarity rate with said first user, as being the ratio of the number of archetypes common with said first user to the total number of archetypes identified from the plurality of experiences, - transmit to said first user information recommending connection with other users having a similarity rate greater than a predetermined similarity threshold.

16. Application according to the preceding claim, characterized in that it further comprises a step for selecting all of the other users from a base of other users, on criteria of geographical proximity with the first user.

17. Application according to any one of the two preceding claims, characterized in that the first user and / or the other users comprise machines and / or robots and / or other intelligent systems such as artificial intelligences (AI).

18. Application according to any one of the three preceding claims, characterized in that the step of transmitting connection recommendation information comprises a vocal communication of this recommendation information to the first user.

19. Application according to any one of the four preceding claims, characterized in that the step of determining the similarity rate comprises a weighting of the different experiences lived and / or imagined by the first user and / or by the other users.

20. Application according to any one of the five preceding claims, characterized in that the experience stories captured from other users are entered by voice recognition.

21. Application according to any one of the six preceding claims, characterized in that the experience stories captured from other users are extracted from a blockchain and decrypted before being processed.

22. Application according to any one of the seven preceding claims, characterized in that it further comprises a step for transmitting to the first user information recommending connection with other users having a similarity rate lower than a predetermined maximum rate.

23. Application according to any one of the eight preceding claims, characterized in that the step of capturing a plurality of experience stories is carried out on a set of first users interacting with the same connected electronic object.

Images