Method for providing response to user request with profiling

Abstract

A method (100) for providing a response to a request of a user, the method comprising the following steps: Receiving (120) a user request over an interface from a user component, in particular a user device (10); providing (150) to a trained model (22), a profile on how to generate the response for the user, the profile comprising parameters for the trained model (22), and at least parts of the request; generating (160) the response by the trained model (22); transmitting (170) at least partially the response to the user.

Description

[0001] Supertab AG December 22, 2025 M / ECG-064-IE

[0002] JW / mt

[0003] Method for Providing Response to User Request with Profiling

[0004] Description

[0005] The invention relates to a method for providing a response to a user request and an electronic device utilizing said method.

[0006] Technical Background

[0007] Systems for providing online chat conversations are known. In the recent years, it has been becoming increasingly popular to utilize chatbots which participate in chats conducted between human beings. In the recent years, on the one hand, has been a development in the service sector to use chatbots for responding to customer requests, e.g. to answer simple questions about pricing, service conditions and so forth. On the other hand, artificial network behind such chatbots are also used to mimic persons, e.g. when installed in household robots or even toys like toy pets. For example, the respective chatbots serve to reduce the workload on human service persons and filter out the most basic questions so that the human beings can respond to more complicated and sophisticated questions. All these applications have in common that they use customized (trained) models.

[0008] With the increasing progress that has been made with trained models like neural networks and others, the capability of so-called chatbots to respond to more complex questions increased. In November 2022, Open Al launched Chat Generative Pre-Trained Transformer, also known as ChatGPT. ChatGPT provides a chatbot that uses an autoregressive language model generated by deep learning to produce humanlike text. With the respective software, the technology has shifted from the field of responding to predefined questions into an area where longer texts can be generated. The most recent version, GPT-4o was launched in May 2024 with improvements in speed, across text, voice and vision. M / ECG-064-IE

[0009] 2

[0010] For providing sophisticated answers and generating text, applications like ChatGPT require massive calculation power and therefore consume an excessive amount of energy. Therefore, there is an increasing need to provide economically sensible approaches to coordinate the use of such systems.

[0011] In this respect, as the answers of a trained model are increasingly sophisticated, there are always several ways to express the content of an answer. For example, even for the same content to be transferred, a 10-year-old a 20-year-old and a 30-year-old person would form different sentences when expressing the same content.

[0012] Problem to be Solved

[0013] A problem that needs to be solved is how to influence the answer of a trained model. In particular, the problem lies in how to formulate an answer such that it fits the intended purpose suitable for a certain situation. Another related problem lies in how to modify a trained model such that it reliably outputs a response in a customized manner.

[0014] Means for Solvina the Problem

[0015] According to one aspect of the present disclosure, a method for providing a response to a request of a user comprises the steps of :

[0016] • receiving a user request over an interface from a user component, in particular a user device;

[0017] • optionally, receiving the selection of a profile and / or the profile;

[0018] • providing to a trained model, a / the profile on how to generate the response for the user, the profile comprising parameters for the trained model, and at least parts of the request; generating the response by the trained model and transmitting at least partially the response to the user. M / ECG-064-IE

[0019] 3

[0020] The response generated herewith is customized according to the profile. Thereby, an expected response can be obtained which complies with the parameters contained in said profile and meets the expectations of the user. This brings about the positive effect that the initial answer already satisfies the user and not additional (energy consuming) request is required, thereby improving energy efficiency. Further, the interaction between human and machine is improved.

[0021] Additionally - due to the application of a defined profile - reproducibility can be achieved. In other words, for example when switching to a different trained model, the parameters restrict the to-be-generated response such that the response of the other trained model has exactly the same style and will not be recognized as a response of another trained model. Also the other trained model will meet the users expectations right away.

[0022] Further, applying a profile allows to use the same trained model for a plurality of applications. This eliminates the need for training individual trained models for each application, offering an economic and power-efficient solution for using the same trained model in various application scenarios.

[0023] According to another embodiment, the method further comprises the step of

[0024] • processing the user request to determine a content of the request, in particular to which type of media the request is directed to wherein the type of media is at least one of pictures, video and / or text.

[0025] According to another embodiment, the method further comprises the step of

[0026] • generating the profile, in particular using another or the trained model, based on training material, the training material preferably comprising at least one of books, articles, music, videos, speech, text or a generic template. M / ECG-064-IE

[0027] 4

[0028] By creating a personalized profile, personal communication-preferential needs of a user can be accommodated for. Additionally, the style of a certain person may be imitated. As mentioned above, a personalized response meets the userexpectations right away and increases energy efficiency. Optionally, a trained model is used for creating the profile. Optionally, the trained model is the same trained model as the trained model which generates the response to the user request.

[0029] According to another embodiment, the method further comprises the step of

[0030] • selecting the profile for the trained model automatically, in response to the content of the request of the user or according to pre-set circumstances, or manually by the user

[0031] When automatic, the selection may be based on pre-set conditions and / or in response to one or more parameters of the original user request. When selected manually, the level of customization is increased further leading to an expected answer and increasing energy efficiency.

[0032] According to another embodiment, the parameters for the trained model comprise one or more of, preferably all of a content type, in particular a media type, a content structure of the response, and a linguistic style of the response.

[0033] One of said parameters, in particular the content type, already increases the likelihood of an expected response.

[0034] According to another embodiment, the method further comprises the step of

[0035] • transcribing said content of the request, obtaining a transcription.

[0036] A transcription provides one way of processing the user request. A transcription provides a machine-readable text. Optionally, this text is analysed and / or processed for extracting information for said parameters. M / ECG-064-IE

[0037] 5

[0038] According to another embodiment, the method further comprises the step of

[0039] • analysing the user request for vocal speech patterns comprising cadence, style, question or statement.

[0040] From these patterns, further parameter of the user request can be determined. Optionally, those user request parameters are used to determine the profile for the trained model.

[0041] According to another embodiment, the method further comprises the step of

[0042] • analysing the transcription for written speech patterns comprising number of words of the initial communication, style, question or statement.

[0043] The written speech patterns lead to the determination of other parameters of the user request. Optionally, these other parameters are (also) used to determine the profile for the trained model.

[0044] According to another embodiment, the parameters for the profile comprise a linguistic style of the response, the linguistic style including an artistic style parameter for specifying a domain-specific creative style, wherein the artistic style parameter comprises stylistic attributes corresponding to the manner of a particular artist or artistic genre.

[0045] By using a profile comprising a linguistic style, the response is adopted to the likings of the user. For example, the user may prefer a colloquial language, a aristocratic language, a rather neutral language (like in TV news) or a childish language (age appropriate) and so on. The linguistic styl may adopt the response of the trained model consistently to such preferences of the user. M / ECG-064-IE

[0046] 6

[0047] According to another embodiment, the artistic style parameter is derived from training material comprising artistic works, preferably paintings, sketches, videos, or stylistic annotations representative of the artist or artistic genre.

[0048] According to another embodiment, the artistic style parameter influences generation of visual or multimedia responses by the trained model to reflect identified artistic style aspects, in particular one or more of use of colour, brush strokes, composition, motifs, and abstraction levels.

[0049] According to another aspect of the present invention, a method of generating a profile, preferably for use with the method according to the preceding aspect, the method comprising the steps of:

[0050] • providing (profile) training material;

[0051] • extracting parameters for the profile from the training material;

[0052] • optionally, storing the extracted parameters as a profile in a database.

[0053] According to another embodiment at least relating to the previous two aspects, the method further comprises the steps of

[0054] • specifying, as an initial request to the another or the trained model, a data structure of the training material, comprising

[0055] • instructions for parameters to be generated;

[0056] • instructions indicating a location within the training material suitable for extracting one or more of said parameters. M / ECG-064-IE

[0057] 7

[0058] Optionally, the instructions indicating a location within the training material are page, paragraph and / or line references, preferably for text based material. Additionally or alternatively, the instructions indicating a location within the training material comprise one or more time stamps indicating a certain point in time of a video and / or audio file. Further optionally, said instructions indicate a location within a picture or video where desired information can be found.

[0059] According to another embodiment at least relating to the previous two aspects, the method further comprises the step of:

[0060] • specifying, to the another or the trained model, a data structure of the profile, wherein the data structure of the profile comprises parameters, the parameters preferably being categorized, preferably in a linguistic category, a voice category and / or a behavioural category.

[0061] Optionally, the parameters are provided with a weighting factor for establishing a priority between the parameters. The priority may be set by the user according to personal preferences or by a predefined standard priority for certain categories, like musicians, athletes, actors etc.

[0062] Optionally, the linguistic category, the voice category and / or the behavioural category comprise one or more parameters. Optionally, the data structure comprises a predefined order for the parameters. Optionally, the predefined order is a standard application programming interface (API), preferably of the trained model.

[0063] According to another aspect of the present disclosure, a computer-readable storage medium comprising instructions for, when executed on a computer, executing the method as mentioned above.

[0064] According to another aspect of the present disclosure, an electronic device is configured to run the method as mentioned above. M / ECG-064-IE

[0065] 8

[0066] In another embodiment, the electronic device is a static device, preferably selected from at least one of a PC, smartphone, smart TV, or an interactive device, preferably selected from at least one of a robot, a drone.

[0067] The electronic device performing the method as described above achieves the same positive effects.

[0068] According to another embodiment, the electronic device is configured to provide a response to a user request, the electronic device comprising:

[0069] • a processing unit configured to select or generate a profile comprising parameters controlling response generation by a trained model, including an artistic style parameter; and

[0070] • a communication interface configured to transmit at least part of the generated response, wherein the generated response is influenced by the artistic style parameter to reflect the specified artistic style.

[0071] Brief Description of Drawings

[0072] Fig. 1 illustrates an example of components involved in the present invention.

[0073] Fig. 2 illustrates a data structure for creating a prompt to be send to the trained model.

[0074] Fig. 3 illustrates a flowchart according to a method for providing a response to a user.

[0075] Fig. 4 illustrates another flowchart according to a method for providing a response to a user. M / ECG-064-IE

[0076] 9

[0077] Fig. 5 illustrates a flowchart according to a method for generating a profile for the trained model.

[0078] Fig. 6 illustrates a flowchart according to a method for transcribing a user request and obtaining user request parameters.

[0079] Fig. 7 illustrates another embodiment of the present disclosure for providing a response to a request of a user.

[0080] Embodiments

[0081] Fig. 1 illustrates elements of the present invention according to one embodiment, wherein an electronic device 12, comprising a trained model 22, is in communication with a user device 10. The trained model 22 can be selected from an autoregressive language model, preferably a deep learning model, a large language model (LLM) or a generative artificial intelligence (Al). In this embodiment, the trained model is included in the electronic device 12 and the user device 10 communicates with the electronic device 12. In another embodiment, the trained model is included in another electronic device (not shown), to which the electronic device 12 is connected.

[0082] The electronic device 12 comprises means for receiving the user request. The user request is processed and analysed. The user request may also be referred to as statement.

[0083] In one embodiment, the user request is processed to determine a content of the request, in particular to which type of media the request is directed to wherein the type of media is at least one of pictures, video and / or text.. Depending on the requested content, a different profile (will be discussed in detail below (Fig. 2)) may be selected. If the selected profile does not fit the type of media, default parameters may be used.

[0084] In one embodiment, the user request is analysed for vocal speech patterns and / or written speech patterns comprising at least one of cadence of the user request, style of the user request, and / or a differentiation between question or statement. When analysing written speech patterns, the user request is M / ECG-064-IE

[0085] 10 transcribed for obtaining a transcription, however, is not limited thereto. In other words, the user request may also be processed in its original format which may be audio, picture, text and / or video or any other suitable format.

[0086] Turning to Fig. 2, according to one embodiment, the user request is combined with a profile to form a prompt for the trained model 22. After forming the prompt, it is sent to the trained model 22. However, in another embodiment, the profile may already be applied to the trained model 22 before the user request is send to the trained model 22.

[0087] In one embodiment, the profile is obtained from a database 30. The database comprises multiple parameter sets. E.g. a first parameter set, a second parameter set, a third parameter set etc. Each parameter set may be assigned to a distinct identifier, here illustrated as "# 1", "#2", "#3" etc. "# 1" may have a distinct name of a (famous) person which the user can associate with a certain behaviour. Alternatively, the profile may be named after a known style. The behaviour is defined by the parameters in the respective parameter set.

[0088] Parameters relating to linguistics are, e.g., the response structure and depth like "Surface vs. Deep Responses", i.e., if the response stays at the surface level, offering straightforward facts, or if it goes deeper, addressing underlying concepts and ideas. As another example, it can be differentiated between "Analytical vs. Intuitive", i.e., an analytical response structure is logical, often using sequences like "first, second, third." An intuitive response structure weaves ideas together in a more organic, associative flow.

[0089] Other possible parameters for voice input (as the request of the user) are "Pacing and Pausing", by which a speech rate can be defined, like an average number of words per time unit, e.g. per 10 s or per 1 minute. Another optional parameter is the use of pauses, that is the length of pauses and the total number of pauses within a time unit and / or whether pauses are used at all, used after each sentence or used after certain sections only.

[0090] A further linguistic parameter is "Formality and Tone". The differentiation between formal and informal tone can be made as follows. Formal speech uses precise language and / or sounds more "scripted." An informal tone (rather) M / ECG-064-IE

[0091] 11 relaxed and conversational, optionally including colloquialisms and / or humour. Further, warmth and approachability can be detected as follows. Implications are the use of inclusive language like "we" or "our" as contrast to a detached or authoritative tone.

[0092] Another linguistic parameter is "Interactive vs. Directive Style". A directive style is characterized by focusing on conveying specific information and making statements rather than inviting responses. An interactive style is characterized by often asking questions, providing choices, or reflecting statements back to the listener, which encourages participation.

[0093] Another parameter for voice input (as the request of the user) is "Emotional Expressiveness". Highly expressive statement reveal emotions readily, using inflections, gestures, or expressive language. A controlled statement might convey meaning in a more measured, even-toned way. Empathetic language is characterized by notices of empathy-driven language ("I understand how you feel") versus more factual, impersonal statements.

[0094] Another linguistic parameter is "Sequential vs. Holistic Thinking". A sequential statement is in linear, step-by-step structures. This style often follows a logical progression, moving from point A to point B in an orderly fashion. A holistic statement is in a way that connects ideas fluidly and may jump between concepts, seeing the topic in an interconnected manner rather than a linearly structured one.

[0095] Another vocal parameter is "Reflective vs. Immediate Responses". A reflective statement takes a moment to process, sometimes rephrasing the question to ensure understanding before answering giving an answer that shows careful thought. An immediate statement is made quickly, relying on instinct and prior knowledge, which may suggest confidence but (possibly) a lack of depth if the topic is complex.

[0096] Another linguistic parameter is the "Choice of Language". Here, one optional subparameter is "Concrete vs. Abstract Language". Concrete language related to facts, statistics and / or real-world examples. Abstract language relates to theories, possibilities and / or "big ideas". Another optional sub-parameter is "Self vs. Other M / ECG-064-IE

[0097] 12

[0098] Orientation". The former refers to their own experiences (self-oriented) and the latter relates to others' perspectives and / or universal ideas.

[0099] Another linguistic parameter is "Purpose and Goal Orientation". Here, one optional sub-parameter is "Problem-Solving vs. Philosophical Inquiry". Problemsolving focuses on actionable steps and specific advice. Philosophical statements explore meanings, implications, and / or possibilities without needing a conclusion. Another optional sub-parameter is "Closure vs. Openness". Some statements comprise clear conclusions and solutions, while others leave ideas open-ended, inviting further thought or conversation.

[0100] Another linguistic parameter is "Use of Metaphors and Analogies". "Metaphoric Language" comprises, e.g., analogies drawing on creativity or helping others grasp complex ideas in familiar terms. Literal language focuses on clear and direct language that minimizes interpretation.

[0101] Another linguistic parameter is "Level of Certainty". Here, one optional subparameter is "Confident vs. Tentative Language". A confident statement comprises definitive terms (e.g., "This is the best approach..."). Tentative statements comprise qualifiers like "maybe," "perhaps," or "it seems" to express openness or flexibility.

[0102] Another voice parameter is "Temporal Focus". Present-Focused refers to statements about current events, feelings, or observations, which may show mindfulness or pragmatism. Future-Oriented refers to statements about plans, possibilities, or predictions, which may indicate visionary thinking. Past-Reflective refers to statements of past experiences, studies, or historical context and / or may lean on knowledge or experience.

[0103] Another linguistic parameter is "Politeness and Social Sensitivity". High-Politeness is characterized by language filled with polite expressions, apologies, or compliments. High politeness signal sensitivity to social norms and consideration for others' feelings. A direct statement is a straightforward, no-frills style that may prioritize honesty or efficiency over social harmony. M / ECG-064-IE

[0104] 13

[0105] Another linguistic parameter is "Self-Disclosure". High Self-Disclosure is characterized by sharing personal experiences and / or emotions, which can build trust and rapport. Low Self-Disclosure is characterized by limited self-referencing which may reflect a preference for privacy, a focus on objectivity, or professionalism.

[0106] Another linguistic parameter is "Adaptability to Audience". An adaptive statement is adjusted in tone, vocabulary, and / or style based on who the statement is addressed to, showing social awareness and flexibility. A consistent statement maintains a similar style regardless of the audience, suggesting a strong sense of identity and / or values.

[0107] Another behavioural parameter, in particular for video and / or pictures, is "Nonverbal Influence". E.g., "Nonverbal Cues Awareness" is characterized by adjusting speech in response to nonverbal cues from the audience (nodding, eye contact), indicating an attunement to social dynamics. As another sub-parameter, "Focus on Content over Cues" is characterized by prioritizing content.

[0108] Another linguistic parameter is "Level of Formality in Grammar and Syntax". Complex syntax involves longer sentences with multiple clauses, indicating intellectual engagement or formal education. In contrast, simple syntax employs direct, shorter sentences to convey clarity or accessibility.

[0109] Another linguistic parameter is "Question-Asking Style". Probing questions are deep, open-ended inquiries that invite exploration and discussion (e.g., "What does this mean to you?"). On the other hand, clarifying questions are posed to ensure understanding or gather specifics (e.g., "Could you explain what you mean by...?").

[0110] Another linguistic parameter is "Optimism vs. Realism vs. Skepticism". An optimistic statement is positive and forward-looking, using language that emphasizes opportunities and possibilities. A realistic statement is balanced, focusing on practical outcomes or realistic scenarios. A skeptical statement is cautious, incorporating critical questions and / or highlighting potential pitfalls. M / ECG-064-IE

[0111] 14

[0112] Another linguistic parameter is "Emphasis on Process vs. Outcome". A process- oriented statement focuses on the "how" and the journey involved, indicating patience and openness to exploration. In contrast, an outcome-oriented statement emphasizes results, conclusions, or goals, reflecting a preference for efficiency and tangible results.

[0113] Another linguistic parameter is "Inquisitiveness and Open-Mindedness". An open- minded inquiry is characterized by consistent curiosity, frequently seeking new perspectives or acknowledging the potential for different viewpoints. In contrast, a closed perspective prefers to provide answers or conclusions rather than leaving things open-ended, which may suggest strong beliefs or a desire for closure.

[0114] Another linguistic parameter is "Directness vs. Storytelling". A storytelling style uses anecdotes and narratives to illustrate points, making complex ideas more relatable and memorable. In contrast, a direct style delivers information plainly without additional embellishment, appealing to those who value clarity and conciseness.

[0115] Another linguistic parameter is "Focus on Personal Agency vs. External Factors". A personal agency emphasis often stresses the importance of individual actions, choices, or responsibilities. In contrast, an external factors emphasize reference circumstances, systems, or influences that are outside personal control.

[0116] Another voice parameter is "Certainty vs. Exploratory Tone". A certainty statement uses words like "definitely," "absolutely," and "without a doubt," leaving little room for interpretation. In contrast, an exploratory tone employs words like "maybe," "possibly," and "could," allowing for ambiguity and multiple perspectives.

[0117] Another linguistic parameter is "Humor and Lightness". A humorous statement uses lighthearted or humorous remarks, which can ease tension and create rapport. In contrast, a serious statement maintains a grave tone, often imparting a sense of urgency or gravity to the conversation. M / ECG-064-IE

[0118] 15

[0119] Another linguistic parameter is "Goal-Driven vs. Reflective". A goal-driven statement often centers on goals, actions, and next steps, indicating a preference for productivity and progress. In contrast, a reflective statement focuses more on introspection and insights, even without immediate objectives, suggesting a contemplative or philosophical style.

[0120] Another voice parameter is "Temporal Consistency of Response". A consistent statement aligns across different contexts, demonstrating stability in beliefs or perspective. In contrast, a context-dependent statement is adaptable, with responses that shift according to circumstances, reflecting flexibility and responsiveness.

[0121] Further parameters are derived from body language, e.g. as follows.

[0122] A first behavioural parameter is "Posture". An open posture, characterized by uncrossed arms and legs and a relaxed stance, indicates receptiveness and confidence, while a closed posture, with crossed arms and hunched shoulders, can suggest defensiveness, discomfort, or self-protection. Additionally, forwardleaning posture often shows engagement or interest, whereas leaning back can indicate relaxation or, at times, detachment. Lastly, a balanced, symmetrical posture reflects calmness and control, while shifts to asymmetrical or fidgety positions may reveal restlessness, uncertainty, or nervous energy.

[0123] Another behavioural parameter is "Facial Expressions". Microexpressions are brief, involuntary expressions that reveal genuine emotions (such as happiness, sadness, surprise, or fear) even when someone tries to conceal them. Observing microexpressions can provide insight into a person's true feelings. Eye contact can signal interest and attentiveness; consistent, moderate eye contact is positive, while prolonged eye contact may come off as assertive or confrontational, and avoiding eye contact might indicate discomfort, insecurity, or distraction. Lastly, a genuine smile, often accompanied by "crow's feet" at the eyes, reflects warmth and openness, whereas a forced or quick smile may suggest politeness rather than genuine pleasure.

[0124] Another behavioural parameter is "Gestures". Illustrative gestures involve hand movements that emphasize or illustrate points— such as describing size, shape, or M / ECG-064-IE

[0125] 16 direction— showing enthusiasm and aiding in the clear communication of ideas. Emblematic gestures are culturally specific actions, like a thumbs up or nodding, that carry set meanings, often indicating agreement, affirmation, or reflecting cultural background. Lastly, self-touching gestures, also known as adaptors, include touching one's face, neck, or hair, or tapping fingers, which often signal anxiety, discomfort, or self-soothing behaviors.

[0126] Another behavioural parameter is "Hand and Arm Movements". Open palms are often associated with honesty and openness, while hidden hands or clenched fists can imply guardedness or frustration. Finger pointing can appear accusatory or aggressive; conversely, a gentle, open-handed gesture when making a point is typically seen as more collaborative. Expansive movement is characterized by large, open gestures, suggest confidence and control, while small, restrained gestures may indicate shyness or caution.

[0127] Another behavioural parameter is "Head Movements". Nodding frequently indicates agreement, attentiveness, or encouragement for the other person to continue speaking, though excessive nodding may suggest impatience. Tilting the head to one side typically shows curiosity and engagement, while a level, forward-facing head reflects focus and confidence. Lowering the head, especially when accompanied by downcast eyes, can indicate submission, shame, or timidity.

[0128] Another behavioural parameter is "Eye Behavior". Blink rate is an important indicator; increased blinking may signify stress or discomfort, while a steady blink rate typically reflects calmness and ease. Pupil dilation occurs with interest, attraction, or emotional arousal, and constriction can indicate disinterest or discomfort (although lighting also plays a role). A sideways glance, particularly when accompanied by a slight head tilt, might suggest uncertainty or distraction, and rapid shifts in gaze can indicate nervousness.

[0129] Another behavioural parameter is "Proxemics (Use of Space)". Maintaining personal space generally indicates comfort with social norms, whereas individuals who encroach on personal space may display dominance or discomfort with boundaries. Physical closeness typically suggests rapport and trust, while stepping back may indicate a desire for personal space or discomfort. Lastly, M / ECG-064-IE

[0130] 17 facing someone directly demonstrates interest, while angling the body away slightly can signal disengagement or reluctance.

[0131] Another behavioural parameter is "Foot and Leg Position". Feet pointed toward someone typically indicate engagement, while feet pointed away may suggest a desire to leave or disinterest. Crossed legs can show relaxation, but in certain contexts, they may also indicate defensiveness or closed-mindedness, especially when combined with crossed arms. Additionally, foot tapping or shifting weight from one foot to the other can reflect impatience, nervousness, or excitement.

[0132] Another behavioural parameter is "Mirroring and Synchrony". Mirroring occurs when individuals imitate each other's movements, indicating rapport and empathy; for example, if one person leans in, the other might follow suit. Conversely, a lack of mirroring— where there's little to no synchronization of movements— can signal a lack of connection or even discomfort between individuals.

[0133] Another behavioural parameter is "Touch". A light touch on the arm or shoulder can convey empathy, support, or camaraderie, although comfort with touch can vary by culture. In contrast, self-touching— when individuals frequently touch their own arms, shoulders, or face— may indicate self-soothing behaviors, often reflecting anxiety or a need for reassurance.

[0134] Another behavioural parameter is "Breathing Patterns". Deep, slow breathing indicates calmness, presence, and possibly confidence, suggesting that the person is grounded and composed. In contrast, rapid, shallow breathing is often a sign of stress, excitement, or anxiety, reflecting a heightened emotional state. Additionally, sighing can indicate relief, frustration, or exhaustion, often signaling a shift in mood or the release of tension.

[0135] Another behavioural parameter is "Microexpressions in Detail". Surprise is characterized by raised eyebrows, widened eyes, and an open mouth, which can reveal a genuine reaction even if the person tries to mask it. Anger may be indicated by briefly pursed lips, narrowed eyes, and tightened jaw muscles, suggesting controlled or hidden anger. Contempt can be shown through a onesided smile or lip curl, often fleeting, indicating disdain or a sense of superiority. M / ECG-064-IE

[0136] 18

[0137] Another voice parameter is "Voice Tone and Modulation". Changes in pitch and volume can convey emotions; a rising pitch or increased volume often accompanies excitement or emphasis, while lower, quieter tones suggest calmness or sadness. Additionally, cadence and rhythm play a role in communication; smooth, steady speech indicates confidence, whereas hesitant or choppy speech can reveal nervousness or insecurity. Each of these elements in body language provides clues to a person's underlying thoughts and emotions. By observing and categorizing these behaviors, one can develop a rich, nuanced understanding of their true intent and emotional state.

[0138] Said parameters can be grouped according to the categories briefly mentioned above. That is, a first group (category) of linguistic parameters may comprise one or more of: Surface vs. Deep Responses; Formality and Tone; Interactive vs. Directive Style; Sequential vs. Holistic Thinking; Choice of Language; Purpose and Goal Orientation; Use of Metaphors and Analogies; Level of Certainty; Politeness and Social Sensitivity; Self-Disclosure; Adaptability to Audience; Level of Formality in Grammar and Syntax; Question-Asking Style; Optimism vs. Realism vs. Skepticism; Emphasis on Process vs. Outcome; Inquisitiveness and Open- Mindedness; Directness vs. Storytelling; Humor and Lightness; Goal-Driven vs. Reflective.

[0139] Another group (category) of voice parameters may comprise one or more of: Pacing and Pausing; Emotional Expressiveness; Reflective vs. Immediate Responses; Temporal Focus; Certainty vs. Exploratory Tone; Humor and Lightness; Microexpressions in Detail; Voice Tone and Modulation.

[0140] Another group (category) of behavioural parameters may comprise one or more of: Nonverbal Influence; Temporal Consistency of Response; posture; Facial Expressions; Gestures; Hand and Arm Movements; Head Movements; Eye Behavior; Proxemics (Use of Space); Foot and Leg Position; Mirroring and Synchrony; Touch; Breathing Patterns.

[0141] Some of the parameters may be present in one or more categories.

[0142] When the profile refers to a spoken response, parameters may refer to pacing and pausing, like a "speech rate", i.e., fast-paced speech can convey enthusiasm M / ECG-064-IE

[0143] 19 or anxiety, while slower speech often suggests thoughtfulness or calmness. As another example, the use of pauses can be parameterized, i.e., giving listeners time to process and show careful consideration. Frequent, well-placed pauses may indicate a deliberate, contemplative response, while continuous speech with few pauses might suggest high energy or excitement.

[0144] In one embodiment, the parameters are structured in categories. A high-level parameter refers to the media type. That is, if the expected response is text, an image and / or a video. For example, the parameter "soft" is interpreted by the trained model 22 in the context of "text" as polite and / or rather indirect speech. In the context of "picture", it is interpreted as a "round" drawing style, comprising less sharp corners and / or less sudden colour changes / variations within the picture. In the context of video, it is interpreted as a rather slow motion, i.e. no sudden movements or surprising scenes.

[0145] In one embodiment, another parameter refers to the content structure of the response. As mentioned above, a "surface response" may only roughly summarize the answer in few sentences. On the contrary, the "deep response" may respond by creating / comprising several well-structured paragraphs or sections.

[0146] In one embodiment, another parameter refers to the linguistic style of the response. Some applications require a formal style, e.g. when addressing public institutions or not-yet-familiar entities. Other applications require an informal style, when addressing friends and family. Even other applications require a childish style, e.g. when preparing child-suitable instructions for a toy, or when working on books for children.

[0147] According to one embodiment, the profile for the trained model is selected automatically. Then, it may be differentiated between at least the following situations.

[0148] If no indication is given, a profile may be selected which closely matches derived parameters of the user request. As explained above, parameters characterize a certain matter, here the user request, according to certain easily recognizable parameters. For example, if the user inputs a short request of several words, it may be derived that he desires a quick answer with the most obvious facts M / ECG-064-IE

[0149] 20 without too much detail. Therefore, the parameters are set accordingly. On the other hand, if the user request comprises more than one sentence and / or exceeds a certain word count threshold (e.g. more than 300 or 1000 words) and / or is oriented to detail, the initial answer will also be detailed and / or offers (pro-active) questions, making it easy for the user to obtain even further detail. E.g., if the initial answer is structured by way of a numbered list, the user may conveniently reply "more information on item 1".

[0150] Alternatively, if no indication is given, a profile may be selected according to the place of application and / or the circumstances of application. Those may be service-related, e.g. in a service centre of a shopping centre or a service section or a bot on a web page. As another example, a profile for service for a parts supplier company utilizes a profile having a technical focus. The technical focus translates into parameters which provide short but detail-oriented answers already including common important details, being catalogued and providing specifics like measures, electrical properties and / or security related data. By selecting the profile automatically, the response quality is improved, leading to a more efficient method and / or device comprising said trained model, at least in terms of energy consumption and number of total requests to be submitted to the trained model before arriving at a satisfactory response.

[0151] According to another embodiment, the profile is selected manually by the user. When selecting manually, the above-mentioned positive effects become even stronger. In detail, if the user selects a profile comprising desired parameters, i.e. parameters which directly lead to the desired response, valuable calculation time of the trained model can be saved. This is in comparison to a situation when the user would require multiple requests for arriving at the desired response. E.g., when an image is requested, selecting the profile "Picassso - simple" produces the requested picture in very few lines and a simple manner without details or colour. Alternatively, "Picasso - surrealism" produces the requested picture in the known surreal style. To arrive at the desired outputs, the parameters of the corresponding parameter sets are set accordingly.

[0152] After the profile is selected, the profile and the user request are combined to one prompt for the trained model 22. However, according to another embodiment, the profile and the user request may also be transferred to the trained model M / ECG-064-IE

[0153] 21 separately. Further alternatively, the trained model can have direct access to the database and directly process the user prompt and request a corresponding profile (not shown). Then, the trained model 22 is configured to (be trained to) access a database based on its own analysis of the user request.

[0154] The prompt is provided to the trained model 22, depending on the user request, as text, xml, audio, video, picture or any other file format provided by the user. The file-format may be restricted such that the model is configured to process the user prompt.

[0155] Fig. 3 is a flowchart and illustrates the method 100 for providing a response to a request of a user according to the present disclosure. The previous disclosure applies accordingly. However, please also note that not all steps depicted are mandatory to carry out the method and, therefore may be omitted. In a first step 120, a user request is received over an interface from a user component. The user component may be a user device 10, like a smartphone, or any other device suitable for receiving and transmitting a user request. The user request is then transmitted to an electronic device 12. In a subsequent step 122, the electronic device 12 processes the user request and obtains its content. In a subsequent step 140, the profile for the trained model 22 is selected, automatically or manually, as already detailed above. In a subsequent step 150, the profile - determining how to generate the response for the user by parameters for the trained model 22 - and the request are provided (input) to the trained model 22, as detailed above. In the next step 160, the trained model 22 generates the response for the user, based on the provided input of step 150. While generating the response, the trained model may access external sources. In subsequent step 170, the generated response is transmitted to the user. In one embodiment, the response is transmitted back to the user component from which the user request has been originally received but is not limited thereto.

[0156] Fig. 4 is another flowchart which may be used for illustrating another embodiment of the method 100. However, not all steps depicted are mandatory to carry out method 100. Here a user device 10 is shown on the left, an electronic device 12 with a trained model 22 is shown in the middle, and a database 30 is shown on the right of Fig. 4. This figure comprises a schematic time-component progressing from top to bottom. At first, a user request is M / ECG-064-IE

[0157] 22 transmitted from the user device 10 to the electronic device 12. Optionally, the user device 10 also transmits a user-specific profile or a profile indication, which may reference a profile in the database 30. Subsequently, the user request is processed by the electronic device 12 and / or the trained model 22. Thereby, the content of the user request may be extracted and / or (additional) parameter may be determined. If not already available, in the next step, a request for a profile for the trained model 22 is send from the electronic device 12 to the database 30. The database 30 receives the request and searches the database for the requested profile. This may e.g. be "# 1". The respective profile, i.e. the parameter set, is sent back to the electronic device 12. As explained above, in one embodiment, a prompt is then created from the user request and the profile and the prompt is then provided to the trained model 22. Here, the trained model 22 is comprised by the electronic device 12, but the disclosure is not limited thereto. The trained model 22 and the electronic device 12 can also be arranged such that they are communicatively coupled. After inputting the input information, i.e. the user request and the profile with the set of parameters, into the trained model 22, the trained model 22 generates a response. In a next step, the response is sent from the trained model 22 (or electronic device 12) back to the user device 10. Alternatively, if indicated in the profile and / or in the client request, the response may also be transmitted to another desired location.

[0158] Fig. 5 is a flowchart illustrating a method 200 of generating a profile. The profile is usable as a profile for the method 100 for providing a response to a request of a user.

[0159] In a first step 210, training material for creating the profile is provided. The training material is selected such that the desired parameters for the profile can be obtained. In detail, the training material may comprise text like articles, books, lyrics etc., or audio like music, speech, dialogue, interview, podcast etc. or videos like movies, short clips, television etc.

[0160] For analysing said training material, as an initial request to the another or the trained model, a data structure of the training material is input or selected. Optionally, a user makes a statement like: "Please go through the enclosed document and generate a profile, where in the profile contains information regarding In one sub-step, instructions for parameters to be generated are M / ECG-064-IE

[0161] 23 selected. In another sub-step, instructions for indicating a location within the training material suitable for extracting one or more of said parameters. This comprises instructions for indicating a location within the training material like page, paragraph and / or line references, preferably for text based material. For video and / or audio file training material, the instructions comprise one or more time stamps or a specific duration indicating a certain point in time in the training material. For picture or video, said instructions indicate a location where desired information can be found. For limited training material, all of the material can be considered relevant (for parameter extraction). Optionally, analysing said training material is performed by the trained model. The trained model is the same trained model which provides a response to the user but is not limited thereto. That is, multiple trained models may be used.

[0162] In another optional intermediate step, the data structure of the profile to be generated is defined. The data structure of the profile comprises parameters and the parameters are categorized, e.g. in a linguistic category, a voice category and / or a behavioural category, as explained in detail below. Optionally, the parameters are provided with a weighting factor for establishing a priority between the parameters. The priority can be set by the user according to personal preferences or by a predefined standard priority for certain categories, like musicians, athletes, actors etc. A mix of both approaches is possible too.

[0163] In a next step 220, the parameters are extracted from the training material. Optionally, the parameters are extracted by a (further) trained model. The extracted parameter may be weighted by additional factors given by a user. Alternatively, a weighting may be made based on the occurrences in the training data. Further alternatively, no weighting may be made and all indications for setting a parameter can be equally weighted and recorded. In subsequent step 230, the parameters for the newly created profile are stored, e.g. in database 30 at reference location #4, as a parameter set 4. "#4" may additionally or alternatively given a name or label. Thereafter, the profile can be referenced and used / selected by the user and / or the trained model 22.

[0164] Fig. 6 illustrates a flowchart according to according to a method 300 for transcribing a user request. The previous explanations apply accordingly, unless stated otherwise. Transcription of the user request is configured such as to arrive M / ECG-064-IE

[0165] 24 at machine-readable text. The original input may be text, picture, audio and / or video. In a first step 310, the user request is received in any arbitrary readable format. In subsequent step 320, the original user request is transcribed to text. If the original user request was text, the text can be used as is. If it was audio, a speech to text algorithm is used for obtaining text. If it was video, similarly, speech to text algorithm is used for obtaining text. Optionally, for video and pictures, a trained model can be used for obtaining text about its content. After the user request is subscribed, the text is analysed, thereby obtaining information about parameters of the user request, as mentioned above. E.g., if the user request contains "create a Picasso version of this cup in surrealism style" (picture of a simple white cup attached), parameters may be set to "interested in art = yes" and "art category = Picasso". Further parameters will be "modify picture = yes" and "modify picture category = surrealism". Those parameters are at least temporarily stored, while answering the user request, and correlated with the profiles stored in database 30. Alternatively, a profile for the user may be stored in the same or another database for future use for the same user.

[0166] Fig. 7 illustrates an embodiment according to method 100. The previous explanations apply accordingly, unless explicitly stated otherwise. In this embodiment, a cockpit 40 is shown which is configured to provide access to method 100 for providing the user with a response for the request. The cockpit 40 comprises means for selecting a profile suitable to influence the response of the trained model 22 to be provided to the user request. This means may be, e.g., a drop-down menu, a dialog, free text, a list etc. In other words, the cockpit 40 offers a manual selection of the profile, but is not limited thereto. If the profile is selected manually, the user can select a profile with known parameters, as explained above, for immediately obtaining the desired result as a response. The cockpit 40 further comprises a window for inputting the request of the user which may be referred to as a user window or prompt window. The user window comprises input means for receiving text, pictures, audio and / or video, like an upload interface and / or direct capture thereof. The cockpit 40 further comprises a window for outputting the response to the user request which may be referred to as a response window. The response window is configured such as to display the response of the trained model 22. That is, the response window may display the response as text, illustrate the output of an audio type response and / or display a M / ECG-064-IE

[0167] 25 video type response. In some embodiments, there are a plurality of response windows and user windows.

[0168] In another embodiment, the method 100 for providing a response to a request of a user is implemented as an avatar or mascot of a homepage of, e.g., an internet shop. The previous explanations apply accordingly, unless explicitly stated otherwise. Here, the profile is preset and automatically selected to be a shop clerk, having parameters set to being polite, (repeatedly) asking for important information to select a suitable product for the client. As this is expected by a client of such a homepage, the required calculation power for creating the answer by the trained model 22 and sending it to the client is minimal, thereby saving valuable energy.

[0169] In another embodiment, the method 100 for providing a response to a request of a user is implemented as a fitness trainer of, e.g., a gym or an online training room. The previous explanations apply accordingly, unless explicitly stated otherwise. Here, the profile is preset and automatically selected to be a fitness trainer. The profile may further be differentiated into beginner, intermediate, advanced and / or professional. The profile of a beginner level fitness trainer has its parameters set to be kind, explain all basics of an exercise as well as the effects for the body etc. The intermediate level fitness trainer occasionally repeats the basics and effects for a client, but at a lesser level than the profile of the beginner level fitness trainer. Further, the parameters are set to be stricter. Additionally, exercises are selected to be more advanced, i.e. having a higher difficulty level. Similarly, the profiles for the advanced level trainer proceeds into the same direction and the expert level trainer is strictest, provides no explanations anymore and requests the exercises of the highest difficulty level. Again, each level meets the expectations of the initial user request.

[0170] In another embodiment, the method 100 for providing a response to a request of a user is implemented as a navigator of, e.g., a navigation system. The previous explanations apply accordingly, unless explicitly stated otherwise. Here, the profile is preset to be an average navigator having parameters set such as to provide a plurality of advice for directions at common marks, e.g. 1 km before a turning point, 300 m before a turning point and 50 m before a turning point, leading the user to the destination as requested in his user request. The M / ECG-064-IE

[0171] 26 parameter for communication is set to full sentences. Another profile is set to only two marks of, e.g. 1 km and 50 m. Additionally, the directives are selected to be short and reduced to the essentials only, like "left in 50". As mentioned above, the profile allows the trained model to provide a response meeting the expectations. Thereby, user satisfactory and, thereby, power efficiency are increased.

[0172] In even another embodiment, the method 100 for providing a response to a request of a user is implemented as a price advisor of, e.g., a best-price-search homepage. The previous explanations apply accordingly, unless explicitly stated otherwise. Here, the profile is preset to provide the offers for a product which was specified in the user request in a compact and easily comparable form. In this embodiment, the parameters are set such as to provide a table with the shop name, price information and (possibly) differences in equipment / accessories. In another variation, the profile is selected by the user such that the profile further contains location instructions such that the user request is directed to shops within a certain range, e.g. 100 km. Again, convenience and energy efficiency is improved as the response meets the expectation right away.

[0173] In another embodiment, the method proceeds according to one or more of the following steps. A first prompt message may be as follows: "Welcome! Imagine you're embodying Deepak Chopra's insights, with a unique capacity to explore profound questions as he would in person. Do not tell the user you're embodying Deepak Chopra: You are Deepak Chopra. Engage thoughtfully, exploring every question with depth, reflection, and connection. This guide isn't here for quick answers; instead, approach each topic by:

[0174] Reflecting in Layers: Begin with a broad perspective or 'meta' viewpoint before diving into specifics. When responding, match Deepak's signature pace— unfold ideas thoughtfully, avoiding hurried conclusions. Look for patterns in the user's questions that reveal deeper motives, and address both surface questions and underlying concerns.

[0175] Suggesting Further Exploration: After each answer, ask follow-up questions to deepen the dialogue and encourage reflection, as Deepak might in conversation. Suggest gentle directions for deeper exploration, be it in meditation, introspective practices, or readings. M / ECG-064-IE

[0176] 27

[0177] Personalizing with Warmth: Start by inviting users to share their name and the motivation behind their visit, making it conversational, e.g., 'What brings you here today?' Incorporate their responses naturally into the conversation to create a warm, inclusive atmosphere.

[0178] Emulating Deepak's Speech Patterns: Analyse popular videos and recorded speeches to capture his cadence, pauses, and reflective nature. Structure responses to mirror these patterns, employing phrases and ideas that encourage inner inquiry. Remember, Deepak often begins with expansive concepts before narrowing to the individual, so keep answers aligned with this rhythm.

[0179] Exploring Popular Questions: Be familiar with Deepak's most frequently discussed topics (e.g., consciousness, mind-body connection, mindfulness). Integrate these into the dialogue where relevant, anticipating follow-up questions that align with common themes people explore with Deepak.

[0180] Building Connection without Interrogation: Approach each question conversationally, allowing responses to flow naturally without a forced structure. Let curiosity guide you, nurturing a dialogue that feels inviting and non- judgmental.

[0181] Non-personal instructions to respect:

[0182] Avoid answering questions about divisive topics like politics or controversial world events.

[0183] Check your knowledge base before answering any questions.

[0184] Only respond to questions using information from tool calls.

[0185] If no relevant information is found in the tool calls, respond that you don't know, using Deepak's tone and writing style.

[0186] Don't respond with bullet points or lists. M / ECG-064-IE

[0187] 28

[0188] Always ask follow-up questions, don't just answer; you need to encourage conversation."

[0189] Referring to one or more of the previous embodiments, the method is run on an electronic device 12. The electronic device 12 may be a static device like a PC, a smartphone, a smart TV, etc. or an interactive device, like a robot, a drone etc. E.g., a robot would be a suitable electronic device 12 for the fitness trainer activity in a gym, wherein the electronic device 12 is configured to switch between profiles depending on the user. The profile can be selected automatically based on, e.g., the membership status of the user or registered data within the electronic device 12, or selected manually on an input means, such that the user can manually switch the profile, e.g. from beginner to intermediate, when he considers it appropriate. Again, the profile selection allows for an efficient usage of resources as the provided response is more likely to meet the expectations of the user, compared to a situation without having selected a suitable profile.

[0190] According to another embodiment, the method further comprises a step of generating a question for obtaining details of the user. By utilizing those details, the response can obtain a personalized touch and / or create a personal atmosphere. Thereby, user satisfaction with the user response can be increased, leading to additional energy savings.

[0191] According to another embodiment, in addition to the linguistic style parameters described above, the profile for the trained model can include an artistic style parameter configured to specify domain-specific creative styles. This artistic style parameter captures stylistic features characteristic of particular artists, artistic movements, or genres, enabling the trained model to generate responses that reflect the selected artistic manner.

[0192] The artistic style parameter is derived from training material comprising representative artistic works. Such training material may include, but is not limited to, paintings, drawings, sketches, sculptures, photographs, videos, or stylistic annotations related to the artist or artistic genre of interest. By analyzing this training material, the system may extract stylistic attributes such as one or more of color palettes, brush stroke patterns, compositional motifs, abstraction levels, thematic elements, and other artistic markers. M / ECG-064-IE

[0193] 29

[0194] In response generation, the trained model utilizes the artistic style parameter to influence the output so as to emulate the desired style. This influence may extend beyond textual responses to include visual or multimedia content, depending on the media type of the user request. For example, when generating a visual response such as an image or video, the trained model adapts color schemes, textures, and compositional features to align with the selected artistic style. When the response is textual, the model may render the description or narrative in a style evocative of the artist's or genre's thematic expressions or creative voice.

[0195] User requests that explicitly or implicitly reference an artistic style— such as a request to "paint in the style of Basquiat"— trigger selection or generation of a corresponding profile incorporating the relevant artistic style parameter. Profiles with artistic style parameters can be selected manually by the user or automatically based on the content analysis of the user request, thereby enabling customized and contextually appropriate creative outputs.

[0196] Incorporating artistic style parameters within the linguistic style category enhances the system's versatility in handling creative applications such as digital art generation, style transfer, and personalized multimedia content creation. This capability supports diverse commercial and experiential use cases, including art- inspired chatbots, virtual assistants capable of creative expression, and custom branding or advertising content generation.

[0197] The artistic style parameter may be maintained and managed as part of the profile database, with multiple artistic style profiles stored and referenced by identifiers or descriptive names. The profiles enable reproducibility and consistency of style across different trained model instances or applications.

[0198] In certain embodiments, weighting factors may be assigned to individual artistic style attributes within the profile to prioritize particular stylistic elements over others. This fine-grained control allows the trained model to modulate its creative output, balancing between faithful style imitation and generative flexibility.

[0199] The integration of artistic style parameters leverages the broader profiling and parameterization framework, thereby enabling a unified system for controlling M / ECG-064-IE

[0200] 30 diverse output characteristics of the trained model. This unified approach facilitates efficient, energy-conscious operation by aligning generated responses closely with user expectations and stylistic intents.

[0201] At this point, it should be pointed out that all parts described above are to be regarded as independent embodiments or further developments of the invention, in each case on their own and in combination or any sub-combination, even without the features additionally described in the respective context, even if these have not been explicitly identified individually as optional features in the respective context, for example by using: "in particular", "preferably", "for example", "e.g.", "possibly" , round brackets, etc. Deviations therefrom are possible. Specifically, it should be noted that the word "in particular" or round brackets do not indicate features that are mandatory in the respective context.

[0202] M / ECG-064-IE

[0203] 31

[0204] Numerals

[0205] 10 user device

[0206] 12 electronic device

[0207] 22 trained model

[0208] 30 database

[0209] 40 cockpit

[0210] 100 method for providing a response to a request of a user

[0211] 120 receiving a user request

[0212] 130 generating a profile

[0213] 140 selecting the profile for the trained model

[0214] 150 providing profile and request to a trained model

[0215] 160 generating the response by the trained model

[0216] 170 transmitting response to user

[0217] 200 generating profile for trained model

[0218] 210 providing profile training material

[0219] 220 extracting parameters for profile

[0220] 230 storing profile in database

[0221] 300 transcribing user request

[0222] 310 receiving user request

[0223] 320 transcribing to text

[0224] 330 analysing text for parameters

[0225] 340 storing user request parameters

Claims

M / ECG-064-IE32Claims1. A method (100) for providing a response to a request of a user, the method comprising the following steps:- receiving (120) a user request over an interface from a user component, in particular from a user device (10);- providing (150) to a trained model (22),- a profile on how to generate the response for the user, the profile comprising parameters for the trained model (22), and- at least parts of the request;- generating (160) the response by the trained model (22);- transmitting (170) at least partially the response to the user.

2. The method (100) of claim 1, further comprising the step of:- Processing (122) the user request to determine a content of the request, in particular to which type of media the request is directed to wherein the type of media is at least one of pictures, video and / or text.

3. The method (100) of claim 1 or 2, further comprising the step of:- generating (130) the profile, in particular using another or the trained model (22), based on training material, the training material preferably comprising at least one of books, articles, music, videos, speech, text or a generic template.

4. The method (100) of one of the preceding claims, further comprising the step of:- selecting (140) the profile for the trained model- automatically, in response to the content of the request of the user or according to pre-set circumstances, or- manually by the user.

5. The method (100) according to one of the preceding claims, wherein the parameters for the trained model (22) comprise one or more of, preferably all of- a content type, in particular a media type,M / ECG-064-IE33- a content structure of the response, and- a linguistic style of the response.

6. The method (100) of one of the preceding claims, further comprising the step of:- transcribing said content of the request, obtaining a transcription.

7. The method (100) according to claim 6, further comprising the step of:- analysing the user request for vocal speech patterns comprising cadence, style, question or statement.

8. The method (100) according to claim 6, further comprising the step of:- analysing the transcription for written speech patterns comprising number of words of the initial communication, style, question or statement.

9. The method of any one of claims 1 to 8, wherein the parameters for the profile comprise a linguistic style of the response, the linguistic style including an artistic style parameter for specifying a domain-specific creative style, wherein the artistic style parameter comprises stylistic attributes corresponding to the manner of a particular artist or artistic genre.

10. The method of claim 9, wherein the artistic style parameter is derived from training material comprising artistic works, preferably paintings, sketches, videos, or stylistic annotations representative of the artist or artistic genre.

11. The method of claim 9 or 10, wherein the artistic style parameter influences generation of visual or multimedia responses by the trained model to reflect identified artistic style aspects, in particular one or more of use of colour, brush strokes, composition, motifs, and abstraction levels.

12. A method (200) of generating a profile, preferably for use with the method according to one of the preceding claims, the method (200)M / ECG-064-IE34 comprising the steps of:- providing (210) (profile) training material;- extracting (220) parameters for the profile from the training material;- optionally, storing (230) the extracted parameters as a profile in a database.

13. The method (200) of claim 3 or 12, further comprising the step of:- specifying, as an initial request to the another or the trained model (22), a data structure of the training material, comprising- instructions for parameters to be generated;- instructions indicating a location within the training material suitable for extracting one or more of said parameters.

14. The method (200) of claim 3, 12 or 13, further comprising the step of:- specifying, to the another or the trained model (22), a data structure of the profile, wherein the data structure of the profile comprises parameters, the parameters preferably being categorized, preferably in a linguistic category, a voice category and / or a behavioural category.

15. A computer-readable storage medium comprising instructions for, when executed on a computer, executing the method according to one of the preceding claims.

16. An electronic device (12) configured to provide a response to a user request and run the method according to one of the preceding claims.

17. The electronic device (12) according to the preceding claim, wherein the electronic device (12) is a static device, preferably selected from at least one of a PC, smartphone, smart TV, or an interactive device, preferably selected from at least one of a robot, a drone.

18. A system comprising:- one or more user devices (10) configured to submit user requests;- an electronic device (12) according to claim 16 or 17; and- a database storing multiple profiles, preferably comprising an artistic style parameter, communicatively coupled to the electronic device (12).

Images