Virtual consumers

Abstract

An apparatus comprises means 12 for training or conditioning a neural network 14 based on a data set 16 to predict consumer sentiment. The data set relates to a sample of real-world target consumers 1

Description

Technical Field The present disclosure relates generally to virtual consumers, and more particularly to virtual consumers predicting the sentiment of real-world consumers towards new products and services. Technical Background A successful product or service provides at least one solution to at least one consumer problem. Commonly, systems for assessing the potential success of new products or services (such as in relation to consumer goods in the manufacturing or retail industry, or films and TV shows in the media and entertainment industry), are based on analyses of consumer reactions to existing products and services in the marketplace. In the manufacturing or retail industry, popularity and success might be judged on the basis of the sales figures and / or aggregated consumer ratings of existing consumer goods or services, whereas a media and entertainment company might judge success by considering the viewing figures and / or aggregated consumer ratings of existing films or TV shows. The success of a new product or service depends on its ability to provide utility to the consumer, typically by solving one or more significant consumer problems. Current systems for assessing the potential success of new products or services often rely on consumer reactions to existing products or services. These systems are inherently limited to evaluating solutions for consumer problems that already have (at least partial) existing solutions. As a result, they often downgrade or overlook the potential of new products that address problems not currently solved by existing products. This limitation arises because these systems, by definition, lack consumer reaction data for such products or services. Consequently, present systems tend to favour the creation of products or services similar to those already available, stifling innovation. This leaves many persistent, diverse consumer problems unaddressed, either partially or entirely. Furthermore, systems that assess consumer preferences among various solutions to their problems are typically based on consumer experiences with existing solutions. This reliance limits consumers' initial expectations of how a novel solution might perform. Consumer knowledge of which solutions best address their problems can be limited until they have actually experienced the new solution. This phenomenon is well-recognized and often described as “I did not think I would like that, but I actually did. ” There is, therefore, a need to provide means to better inform the development of new products and services which takes into consideration all consumer problems and not just those for which solutions or partial solutions are available, and also recognises that consumer solution preferences provided before a solution is experienced may not accurately align with consumer solution preferences after consumers have experienced how well a solution addresses their problems. Summary of the Disclosure According to a first aspect of the present disclosure, there is provided an apparatus comprising: means for training or conditioning a neural network based on a data set, wherein the data set relates to a sample of real-world target consumers, wherein the data set includes identified individuals or clusters in the sample of real-world target consumers, wherein each cluster is a group of people within the sample of real-world target consumers having at least one problem in common, wherein the at least one problem is associated with possible solutions to the at least one problem based on responses from the sample of real-world target consumers, wherein for each identified individual, at least one problem of the individual is associated with possible solutions to the at least one problem based on responses from the sample of real-world target consumers; and means for providing by the trained or conditioned neural network a predicted sentiment of an individual target consumer or of one or more of the identified clusters towards a product or service as a solution or combination of solutions to the at least one problem based on the association between the at least one problem and each of the possible solutions to the problem. Based on the predicted sentiment by the trained or conditioned neural network of an individual target consumer or of one or more of the identified clusters of target consumers towards a product or service as a solution to the at least one problem, examples of the disclosure provide means to better inform the development of new products. The predicted sentiment provides a demonstrably accurate simulation of the sentiment of real-world target consumers and can thus accurately predict whether a proposed new product or service will provide a successful solution or set of solutions to a problem or problems of a particular individual consumer, cluster or clusters, and thus whether that proposed new product or service will be well received by the real-world consumer, cluster or clusters (i.e., how successful the product or service is likely to be). The predicted sentiment may be expressed quantitatively as an output by the trained or conditioned neural network based on a numerical scale. In other examples, the predicted sentiment may be expressed qualitatively as an output by the trained or conditioned neural network based on a natural language response. Possibly, the predicted sentiment provided by the trained or conditioned neural network is an output provided by asking the trained or conditioned neural network as an input one or more questions in relation to the product or service. Possibly, the predicted sentiment provided by the trained or conditioned neural network is an output based on a distribution of sentiment responses, wherein the distribution is based on asking the trained or conditioned neural network as an input the one or more questions in relation to the product or service repeatedly in an automated process. Possibly, each cluster is a group of people within the sample of real-world target consumers having a plurality of problems in common, wherein each of the plurality of problems is associated with possible solutions to those problem based on responses from the sample of real-world target consumers. The product or service may be in relation to consumer goods. The product or service may be in relation to media and entertainment content. The association between the at least one problem and each of the possible solutions to the at least one problem may be quantified by at least one numerical value according to a numerical scale. In some examples, the association between the at least one problem and each of the possible solutions to the at least one problem is defined by the intensity and the degree to which the at least one problem or a plurality of problems of an individual or individuals in a cluster are solved by existing available solutions. Possibly, the at least one numerical value assigned to the association between the at least one problem and each of the possible solutions to the at least one problem is based on a Problem Significance Score in relation to the at least one problem and / or is based on a Solution Preference Score in relation to each solution. Possibly, the Problem Significance Score is a function of a Problem Intensity Score and a Solution Effectiveness Score each assigned by or to the sample of real-world target consumers in relation to the at least one problem, wherein the Problem Intensity Score provides an indication of the degree of intensity of the at least one problem, and the Solution Effectiveness Score provides an indication of how effectively existing available solutions solve the at least one problem. Possibly, people may be grouped into clusters according to their Problem Significance Scores, in addition to having at least one problem in common. Possibly, people may be grouped into clusters according to their Problem Significance Scores which are indicative of at least one problem having a high score and / or at least one problem having a low score. Possibly, a Cluster Problem Significance Score is determined for each identified cluster based on the Problem Significance Scores of each person in that cluster, wherein the Cluster Problem Significance Score for each identified cluster defines a cluster profile which enables the trained or conditioned neural network to predict the sentiment of each of the identified clusters towards a product or service as at least one solution to the at least one problem as a collective. Possibly, the Solution Preference Score is a function of a Solution Intensity of Preference Score and a Problem Set Correlation Score each assigned by or to the sample of real-world target consumers in relation to each solution, wherein the Solution Intensity of Preference Score provides an indication of the degree to which a solution is preferred, and the Problem Set Correlation Score provides an indication of how closely a solution correlates with the at least one problem or plurality of problems associated with the individual or cluster. The Problem Set Correlation Score is designed to impact the Solution Preference Score by adjusting the Solution Intensity of Preference Score (which is more likely based on consumer preferences over existing solutions) by a factor representing how well the solution solves the at least one problem or problems of interest to the individual or cluster, thereby approximating how a consumer would feel about their solution preferences after having experienced the efficacy of the solution. The intensity of solution preference and degree to which any of the plurality of solutions solve the at least one problem or plurality of problems that define an individual or cluster may be quantified by at least one numerical value according to a numerical scale. Possibly, a Cluster Solution Preference Score is determined for each identified cluster based on the Solution Preference Scores of each person in that cluster, wherein the Cluster Solution Preference Score for each identified cluster defines a cluster profile which enables the trained or conditioned neural network to predict the sentiment of each of the identified clusters towards a product or service as at least one solution to the at least one problem as a collective. Training or conditioning the neural network may be by providing as an input to the neural network respective numerical values relating to the association between the at least one problem and the possible solutions to the at least one problem. Accordingly, training or conditioning the neural network may be achieved by providing as an input to the neural network respective numerical values relating both to the problem and solution preferences of a target consumer or clusters thereof. The at least one problem may be based on at least one characteristic associated with each respective identified individual or cluster. The at least one characteristic may include at least one of: demographic information, psychographic information, behavioural traits, and attitude. The clusters may be identified by mathematical, algorithmic or machine learned clustering methods such as, but not limited to, K-means clustering or hierarchical clustering. Cluster Problem Significance and Cluster Solution Preferences may provide intra cluster comparisons and inter cluster comparisons across problems and solutions. The neural network may be a Large Language Model (LLM) or a World Model (WM). Possibly, the apparatus further comprises means for acquiring the data set, wherein the data set is acquired by data collection surveys on the sample of real-world target consumers and / or through observing behaviour of the sample of real-world target consumers. Possibly, the apparatus further comprises means fortraining or conditioning the neural network based on a second data set, wherein the second data set provides the sentiment towards a range of existing or prototype relevant products or services providing at least one relevant solution to at least one relevant consumer problem from another sample of real-world target consumers representative of an individual or one or more of the clusters. The trained or conditioned neural network may provide virtual consumers. Accordingly, the trained or conditioned neural network may predict product or service sentiment responses as an individual virtual consumer or a virtual consumer cluster. According to a second aspect of the present disclosure, there is provided a method comprising: training or conditioning a neural network based on a data set, wherein the data set relates to a sample of real-world target consumers, wherein the data set includes identified individuals or clusters in the sample of real-world target consumers, wherein each cluster is a group of people within the sample of real-world target consumers having at least one problem in common, wherein the at least one problem is associated with possible solutions to the at least one problem based on responses from the sample of real-world target consumers, wherein for each identified individual, at least one problem of the individual is associated with possible solutions to the at least one problem based on responses from the sample of real-world target consumers; and providing by the trained or conditioned neural network a predicted sentiment of an individual target consumer or of one or more of the identified clusters towards a product or service as a solution or combination of solutions to the at least one problem based on the association between the at least one problem and each of the possible solutions to the problem. According to a third aspect of the present disclosure, there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform at least: training or conditioning a neural network based on a data set, wherein the data set relates to a sample of real-world target consumers, wherein the data set includes identified individuals or clusters in the sample of real-world target consumers, wherein each cluster is a group of people within the sample of real-world target consumers having at least one problem in common, wherein the at least one problem is associated with possible solutions to the at least one problem based on responses from the sample of real-world target consumers, wherein for each identified individual, at least one problem of the individual is associated with possible solutions to the at least one problem based on responses from the sample of real-world target consumers; and providing by the trained or conditioned neural network a predicted sentiment of an individual target consumer or of one or more of the identified clusters towards a product or service as a solution or combination of solutions to the at least one problem based on the association between the at least one problem and each of the possible solutions to the problem. Brief Description of the Drawings Some examples will now be described with reference to the accompanying drawing in which: Figure lisa diagrammatic view of an apparatus according to examples of the disclosure. Detailed Description The present disclosure will now be described by way of example only and with reference to the accompanying drawing. Referring to Figure 1, examples of the disclosure provide an apparatus 10. The apparatus 10 comprises means 12 for training or conditioning a neural network 14 based on a data set 16. The data set 16 relates to a sample of real-world target consumers 18. In some examples, the neural network 14 may be a Large Language Model (LLM) or World Model (WM). In examples of the disclosure, training a neural network 14 involves teaching the model to understand and generate language by adjusting its parameters through exposure to vast amounts of text data using learning algorithms. Conditioning a neural network 14 involves guiding a pre-trained model to produce specific responses by providing contextual input prompts, without altering the model's underlying parameters. In examples of the disclosure, a neural network 14 is an artificial neural network. In some examples, the apparatus 10 further comprises means 28 for acquiring the data set 16. The data set 16 may be acquired by data collection surveys on the sample of real-world target consumers 18 and / or through observed behaviour of the sample of real-world target consumers 18. A sample of real-world target consumers 18 may be target consumers of goods or services, or target consumers of media and entertainment content. Target consumers of media and entertainment content can also be considered a target audience. A sample of real-world target consumers 18 may be a representative sample of people large enough and diverse enough to represent a market with a chosen degree of statistical confidence and error. Possibly, the sample of real-world target consumers 18 could be a very large number in the order of millions or billions. Indeed, the sample 18 could be constructed of every single real-world consumer in a market. The sample of real-world target consumers 18 defines a population. Consumer goods are products bought for consumption or use by consumers. Also called final goods, consumer goods are typically the end result of a production or manufacturing process. Clothing, appliances, and food products are examples of common consumer goods. Media and entertainment content include (but are not limited to): film, print, radio, and television. More particularly, media and entertainment content include (but are not limited to): movies, TV shows, radio shows, news, music, podcasts, and written content such as newspapers, magazines, graphic novels, comics, and books. Further examples include journalism and advertisements. Content includes any visual, audio, and audio-visual material. A service is an intangible activity, which also aims at satisfying the needs and wants of a consumer. Services are intangible in nature, i.e., they are non-physical objects that cannot be seen, felt or touched, but can be experienced by the consumer. The ownership of services cannot be transferred from one person to another. They cannot be produced as they are performed as and when required by the customer. Example types of services include business services, social services and personal services. Business services may further be classified as Banking, Insurance, Warehousing, Transportation and Communication. The data set 16 includes identified individuals 19 or clusters 20 in the sample of real-world target consumers 18. In the illustrated example, four such identified individuals 19 or clusters 20 are shown. In practice, the data set 16 may include any number of identified individuals 19 or clusters 20 in the sample of real-world target consumers 18. In a sample of 1500 to 2000 people 22, five or six clusters 20 may be identified. For large samples of 1000000 or more people 22, thousands of clusters 20 may be identified. Each cluster 20 is a group of people 22 within the sample of real-world target consumers 18 having at least one problem 34 (from a wide set of possible problems consumers would like to solve) in common. Accordingly, groups of people 22 are segmented into clusters 20. The at least one problem 34 is associated with possible solutions 36 to the at least one problem 34 based on responses from, or observations of, the sample of real-world target consumers 18. In some examples, each cluster 20 is associated with at least one problem 34 unique to the group of people 22 in that cluster 20. In some examples, each cluster 20 is a group of people 22 within the sample of real-world target consumers 18 having a plurality of problems 34 in common. Each of the plurality of problems 34 is associated with possible solutions 36 to the problem based on responses from, or observations of, the sample of real-world target consumers 18. There can be any number of possible solutions 36 to a problem 34. For each identified individual 19, at least one problem 34 of the individual 19 is associated with possible solutions 36 to the at least one problem 34 based on responses from the sample of real-world target consumers 18. Each identified individual 19 may have a plurality of problems 34 each of which is associated with possible solutions 36. Regarding a consumer good, such as a snack food product, example problems 34 (and associated goals) real-world target consumers 18 may have include (but are not limited to): my mood is too low (lift my mood), I don’t have enough treats or rewards (obtain a treat or reward), my muscle mass is too low (build muscle mass), my dental health is at risk (maintain healthy teeth), and / or my consumption habits are harmful to the environment (reduce environmental impact) Possible example solutions 36 to such problems 34 include (but are not limited to): puffed corn, spicy flavours, compostable packaging, microbiome replenishing ingredients, and / or high protein ingredients. Concerning media and entertainment content, such as Film and TV shows, problems 34 (and associated goals) real-world target consumers 18 (i.e., audiences) may have include (but are not limited to): needing to decompress after a stressful day (want to relax and unwind), craving exhilarating action and thrills (want to experience excitement), feeling isolated or lonely (find connection and understanding), lack of self-understanding (want to gain insight into human nature), and desiring to educate oneself about something new (want to learn and grow intellectually). Possible example solutions 36 to such problems 34 include (but are not limited to): horror films, inspiring films, half hour TV shows, tragic plots, and / or rebellious characters. Accordingly, possible solutions 36 relate to preferences, for example, about genre, plot, protagonist, or other aspects of media and entertainment consumption. In some examples, the at least one problem may be based on at least one characteristic 38, or a plurality of characteristics 38, associated with each identified individual 19 or each respective identified cluster 20. The characteristics 38 include (but are not limited to) at least one of: demographic information, psychographic information, behavioural traits, and attitude. Specific examples, include (but are not limited to): age, income, personality type, attitude to price, and / or frequency of purchasing a product or viewing content. A characteristic 38 can be considered a trait. Characteristics 38 (i.e., traits) may be acquired as an objective definitive (for example, age, income) or as subjective scores (for example, rating attitudes to healthy living on a numerical scale from 0 to 10). The association between problems 34 and possible solutions 36 in relation to under-represented consumer can be identified by examples of the disclosure, for instance, when characteristics 38 are taken into account. The apparatus 10 further comprises means 24 for providing by the trained or conditioned neural network 14 a predicted sentiment 40 of an individual target consumer 19 or of one or more of the identified clusters 20 towards a product or service 26 as a solution or combination of solutions 36 to the at least one problem 34 based on the association between the at least one problem 34 and each of the possible solutions 36 to the problem 34, which may take into account the degree to which the product or service 26 addresses the at least one problem 34. A product or service 26 can be any commercial or creative offering in relation to consumer goods or media and entertainment content, for instance, as described in the examples above. The product 26, need not be a finalised product or service, that is, the product or service 26 can be a prototype, sample, or intermediate in the development of a final product or service. For example, in relation to media and entertainment content, the product or service 26 could be, for instance, a script, an extract of a script, a fancier, a poster, a pitch deck, or a short film. Based on the predicted sentiment 40 by the trained or conditioned neural network 14 of an individual 19 or one or more of the identified clusters 20 towards a product or service 26 as a solution (or combination of solutions 36) to the at least one problem 34, examples of the disclosure provide means to better inform the development of new products or services. The predicted sentiment 40 provides a demonstrably accurate simulation of the sentiment of real-world target consumers 18 towards the product or service 26 and can thus accurately predict whether a proposed new product or service 26 will provide a successful solution or combination of solutions 36 to a problem 34 or problems 34 of an individual 19, a particular cluster 20 or clusters 20, and thus whether that proposed new product or service 26 will be well received by real-world consumers representative of the individual 19, cluster 20 or clusters 20. Examples of the disclosure also foster innovation in relation to new products by identifying new areas for product or service development. Specifically, in relation to media and entertainment content, examples of the disclosure increase the efficiency and effectiveness of content creation at all levels of the industry. In some examples, the predicted sentiment 40 is expressed quantitatively as an output by the trained or conditioned neural network 14 based on a numerical scale, e.g., a Likert Scale. For example, the predicted sentiment 40 may be a score out of 10, e.g., 8 out of 10. Accordingly, each point on the numerical scale corresponds to a different level of predicted preference of the an individual 19 or one or more identified clusters 20 towards the product or service 26. In other examples, the predicted sentiment 40 is expressed qualitatively as an output by the trained or conditioned neural network 14 based on a natural language response. For example, the predicted sentiment 40 may be a descriptive word or phrase, or a description of perceived strengths and / or weaknesses. In other examples, the predicted sentiment 40 is expressed both quantitatively and qualitatively as an output by the trained or conditioned neural network 14. The predicted sentiment 40 constitutes a simulated attitude, thought, judgment, opinion, specific view and / or notion of the trained or conditioned neural network 14. For example, predicted sentiment 40 is an expression of preference towards a product or service 26 as a solution 36 or combination of solutions 36 to the at least one problem 34. The predicted sentiment 40 may indicate, for example, levels of preference ranging from very highly preferred to extremely unpreferred. The predicted sentiment 40 provided by the trained or conditioned neural network 14 is an output provided by asking the trained or conditioned neural network 14 as an input one or more questions in relation to the product 26. The one or more questions in relation to the product 26 may be asked based on natural language. The predicted sentiment 40 is from the perspective of the identified individual 19, cluster 20 or clusters 20 to which, as an input, the one or more questions in relation to the product or service 26 have been asked. In some examples, the predicted sentiment 40 provided by the trained or conditioned neural network 14 as an output is based on what is characteristic of a distribution. The distribution is based on asking the trained or conditioned neural network 14, as an input, the one or more questions in relation to the product or service 26 multiple times in an automated process. For example, the predicted sentiment 40 may be expressed quantitatively as a score of 8 out of 10. The score of 8 out of 10 is characteristic of the distribution, e.g., the highest point on a bell curve when the distribution is classified as normal. Or, for example, the predicted sentiment 40 may be express quantitatively as a chart representing a frequency distribution of sentiments across a scale out of 10. In some examples, the association between the at least one problem 34 and each of the possible solutions 36 to the at least one problem 34 is quantified by at least one numerical value according to a numerical scale. In some examples, the association between the at least one problem 34 and each of the possible solutions 36 to the at least one problem 34 is defined by the intensity and the degree to which the at least one problem 34 (or a plurality of problems 34) of an individual or individuals in a cluster 20 are solved by existing available solutions 36. Accordingly, training or conditioning of the neural network 14 occurs by providing as an input to the neural network 14 respective numerical values relating to the association between the at least one problem 34 and the possible solutions 36 to the at least one problem 34, e.g., as defined by the intensity and the degree to which the at least one problem 34 of an individual or individuals in a cluster 20 is solved by existing available solutions 36. In some examples, the at least one numerical value assigned to the association between the at least one problem 34 and each of the possible solutions 36 to the at least one problem 34 is based on a Problem Significance Score in relation to the at least one problem 34 and / or is based on a Solution Preference Score in relation to each solution 36. The Problem Significance Score is a function of a Problem Intensity Score and a Solution Effectiveness Score each assigned by or to the sample of real-world target consumers 18 in relation to each problem 34. The Problem Intensity Score provides an indication of the degree of intensity of the problem 34. The Solution Effectiveness Score provides an indication of how effectively existing available solutions 36 solve each problem 34. The Problem Intensity Score might run from -5 (Extremely low intensity) to +5 (Extremely high intensity). The Solution Effectiveness Score might run from -5 (Extremely low effectiveness) to +5 (Extremely high effectiveness). The respective Problem Intensity and Solution Effectiveness scores are obtained by data collection surveys on the sample of real-world target consumers 18 and / or collected through observation of the behaviour of the sample of real-world target consumers 18. In some examples, a Problem Significance Score = Problem Intensity Score x 2 x Weight1 + (Problem Intensity Score - Solution Effectiveness Score) x Weight2. Where: Problem Intensity Score = -5 to +5 Solution Effectiveness Score = -5 to +5 Weight1 = 0 to 1 Weight1 + Weight2 = 1 This creates a Problem Significance Score for each problem 34 running from -10 to +10. In some examples, the people 22 may be grouped into clusters 20 (i.e., segmented into groups) according to shared pattern of Problem Significance Scores, in addition to having at least one problem 34 in common. The clusters 20 may be identified by a mathematical, algorithmic, machine learned clustering method, such as a K-means or Hierarchical clustering method. The data set 14 is thus analysed to identify clusters 20 with at least one problem 34 in common taking account shared priorities. This can approximate to people 22 in each respective clusters 20 having a set of similar problems 34. The people 22 in each of the identified clusters 20 can be grouped according to Problem Significance Scores which are indicative of at least one of the possible problems 34 being strongly characteristic of each cluster 20. Accordingly, for each respondent and each problem 34, a numerical method can be used to prioritise those problems 34 that are both intensely felt and comparative unsolved by existing products. In some examples, a Cluster Problem Significance Score is determined for each identified cluster 20 based on the Problem Significance Scores of each person in that cluster 20. The Cluster Problem Significance Score for each identified cluster 20 defines a cluster profile which enables the trained or conditioned network 14 to predict the sentiment 40 of each of the identified clusters 20 towards a product 26 based on how well it addresses the at least one problem 34 as a collective set of consumers. Accordingly, once consumers have been grouped into clusters 20 based on their patterns of Problem Significance Scores, the nature of each cluster 20 can be defined, for instance, numerically. Cluster Problem Significance Scores can be calculated across all problems 34. In some examples, cluster profiles can be based on defined high problem significance (intuitively: to understand problems 34 they care about most) and also low problem significance (intuitively: to understand problems 34 they care about least) to define the highest / lowest priorities of those consumers. In some examples, problem significance combination enables intra cluster comparisons and inter cluster comparisons. Examples of the disclosure thus transform complex, unstructured data (e.g., resulting from survey and / or observation) into clusters 20 characterised by quantified Problem Significance Scores and / or Cluster Problem Significance Scores. In examples of the disclosure, an algorithm is used for the transformation of complex, unstructured data (e.g., resulting from survey and / or observation) into clusters 20 characterised by quantified Problem Significance Scores and / or Cluster Problem Significance Scores. In some examples, the intensity of solution preference and degree to which any of the plurality of solutions 36 solve the plurality of problems 34 that define an individual 19 or cluster 20 may be quantified by at least one numerical value according to a numerical scale. Accordingly, training or conditioning of the neural network 14 occurs by providing as an input to the neural network 14 respective numerical values relating to the intensity of solution preference and degree to which any of the plurality of solutions 36 solve the plurality of problems 34. The Solution Preference Score is a function of a Solution Intensity of Preference Score and a Problem Set Correlation Score each assigned by or to the sample of real-world target consumers 18 in relation to each problem 34. The Solution Intensity of Preference Score provides an indication of the degree of intensity to which the solutions 36 are preferred. The Problem Set Correlation Score provides an indication of how each solution 36 correlates with the problem or plurality of problems 34 associated with each individual 19 or cluster 20. The Solution Intensity of Preference Score might run from -5 (Extremely low intensity) to +5 (Extremely high intensity). The Problem Set Correlation Score might range from 0 (no correlation between solution and problem set) to 1 (strong correlation between solution and problem set). The respective Solution Intensity of Preference and Problem Set Correlation Scores are obtained by data collection surveys on the sample of real-world target consumers 18 and / or collected through observation of the behaviour of the sample of real-world target consumers 18 and / or objective analysis of the degree to which a particular solution 36 might solve a particular problem 34. In some examples, a Solution Preference Score = (Solution Intensity of Preference Score x Weight1) + (Problem Set Correlation Score x Solution Intensity of Preference Score x Weight2) Where: Solution Intensity of Preference Score = -5 to +5 Problem Set Correlation Score = 0 to 1 Weight1 = 0 to 1 Weight1 + Weight2 = 1 This creates a Solution Preference Score for each solution 36 running from -10 to +10. Accordingly, adjustments are made to ensure that associations are prioritised in a postexperience state, rather than a pre-experience state. This is valuable because consumers typically don’t know exactly what they want until they experience a solution 36 that addresses a problem 34 they have. To ensure solution preference are represented in the post-experience state, respondent data based on pre-experience states data is adjusted to more accurately portray their post-experience state. The adjustment is calculated according to how well a possible solution 36 (e.g., a commercial or creative offering) correlates with the underlying problem(s) 34 of an individual 19 or a cluster 20 representing a particular consumer cluster. This provides a critical mechanism to adjust cluster respondent Solution Intensity of Preference Scores, driven by pre-experiential states, into post-experiential Solution Preference Scores, driven by the degree to which each possible solution 36 might solve the, or each, problem 34, which could (in essence) be surprising for a respondent. This captures the well-recognised phenomenon of “I did not think I would like that, but I actually did”. In some examples, a Cluster Solution Preference Score is determined for each identified cluster 20 based on the Solution Preference Scores of each person in that cluster 20. The Cluster Solution Preference Score for each identified cluster 20 defines a cluster profile which enables the trained or conditioned network 14 to predict the sentiment 40 of each of the identified clusters 20 towards a product or service 26 based on how well a solution 36 addresses the at least one problem 34 as a collective set of consumers. Accordingly, once consumers have been grouped into clusters 20 based on their patterns of Problem Preference Scores, the nature of the post-experience solution preferences for each cluster 20 can be defined, for instance, numerically. Cluster Solution Preference Scores can be calculated across all solutions 36. In some examples, the Problem Significance and Solution Preference Scores and Cluster Problem Significance and Solution Preference Scores correspond to a set of ordinal, single peaked preference relations, which can be used to train or condition the neural network 14. In some examples, the apparatus 10 further comprises means 30 for training or conditioning the neural network 14 based on a second data set 32. The second data set 32 provides the sentiment 42 towards a range of existing or prototype relevant products or services providing at least one relevant solution to at least one relevant consumer problem or service from another sample of real-world target consumers representative of an individual 19 one or more of the clusters 20. In examples of the disclosure, the trained or conditioned neural network 14 provides product or service sentiment responses as virtual consumers or virtual consumer clusters which predict the sentiment of real-world consumers towards new products or services 26. The trained or conditioned network 14 therefore predicts the impact of new products or service 26 on real-world consumers via a virtual version of such consumers. In some examples, Python software is associated with means 12 for training the neural network 14 and / or means 24 for providing by the trained or conditioned neural network 14 a predicted sentiment 40. For example, Python software may collate profiles, group profiles, specific questions of interest in relation to a product 26 (e.g., posed by a content creator), details of the product 26 etc. Prompt templates may be utilised for use with the neural network 14 (e.g., Large Language Models or World Models). The apparatus 10 according to examples of the disclosure provides an Artificial Intelligence (AI) system based on the neural network 14 using learning approaches such as supervised learning, unsupervised learning, and reinforcement learning or using conditioning approaches such as fine-tuning or prompting. With regard to supervised learning, the AI system is trained using labelled input data, which includes both the unidimensional and multidimensional problem 34 / possible solution 36 scores and distributions, along with the corresponding group labels. The system learns to map the input features to the correct group labels, enabling it to predict the reactions of a specific group (i.e., cluster 20) when presented with new multidimensional offerings. Examples of the disclosure cover all forms of machine learning when applied to the problem of simulating virtual consumers including but not limited to supervised learning, unsupervised learning and reinforcement learning Concerning unsupervised learning, the AI system is trained using unlabelled input data, consisting of the unidimensional and multidimensional problem 34 / possible solution 36 scores and distributions, without explicit group labels. The system learns to identify patterns and structures within the data, such as clusters or latent representations, which can be used to infer the reactions of different consumer or audience groups (i.e., clusters 20) to new multidimensional offerings. Regarding reinforcement learning, the AI system learns through interaction with a simulated environment, where it receives rewards or penalties based on its actions. In this context, the system could be trained to generate or select multidimensional offerings that maximize the predicted satisfaction or engagement of a specific consumer or audience group (i.e., clusters 20), based on their unidimensional problem 34 / possible solution 36 priorities and distinguishing features. The system iteratively refines its strategies to optimize the expected rewards over time. Regarding fine-tuning, a pre-trained neural network 14 is conditioned through further training on a smaller, domain-specific dataset to adapt it to a particular task or problem. In the context of the AI system described, fine-tuning involves taking an AI model that has already been trained on a broad set of data and refining it with specific examples of unidimensional and multidimensional problem 34 and possible solution 36 scores and distributions relevant to a particular consumer group (i.e., cluster 20). This additional training helps the model learn the nuances and specific preferences of the target group, enhancing its ability to predict their reactions to new offerings. Fine-tuning leverages the foundational knowledge acquired during the initial training phase while honing in on the specific characteristics and needs of the target market, thereby improving the model's accuracy and relevance in predicting product success. Regarding prompting, a pre-trained model (for example, a Large Language Model) is conditioned through provision of specific prompts or queries to elicit desired outputs without further training. In the context of the AI system described, prompting involves presenting the model with specific scenarios or descriptions related to the unidimensional and multidimensional problem 34 / possible solution 36 scores and distributions. The AI system uses its pre-existing knowledge to generate predictions or insights based on these prompts. For example, the system might be given a description of a new product feature and asked to predict how a specific consumer group (i.e., cluster 20) would react. Prompting allows the model to utilize its extensive training data to provide immediate, contextually relevant responses, making it a versatile tool for exploring different hypothetical scenarios and their potential impacts on consumer satisfaction and engagement. In operation, the AI system according to examples of the disclosure has wide applications. For example, the AI system has consumer industry applications. Such applications include: assessing and predicting consumer reaction to a product in development; assessing and predicting how product design and messaging will work with different target consumer groups; and identifying production solutions for gaps in the market where underserved problems (also called unmet consumer needs) are currently common. For example, the AI system has audience industry applications. Such applications include: assessing and predicting film &TV script potential during development; assessing and predicting video content for audience attention; drop-off and completion rate as the video proceeds; assessing and predicting advertising content impact on a target audience group; and assessing the nature of the audience most likely to respond well to a piece of media content. The Figure also illustrates a method. The method comprises training or conditioning a neural network 14 based on a data set 16. The data set 16 relates to a sample of real-world target consumers 18. The data set 16 includes identified individuals 19 or clusters 20 in the sample of real-world target consumers. Each cluster 20 is a group of people 22 within the sample of real-world target consumers 18 having at least one problem 34 in common. The at least one problem 34 is associated with possible solutions 36 to the at least one problem 34 based on responses from the sample of real-world target consumers 18. For each identified individual 19, at least one problem 34 of the individual 19 is associated with possible solutions 36 to the at least one problem 34 based on responses from the sample of real-world target consumers 18. The method further comprises providing by the trained or conditioned neural network 14 a predicted sentiment 40 of individual target consumer 19 or of one or more of the identified clusters 20 towards a product 26 as a solution 36 to the at least one problem 34 based on the association between the at least one problem 34 and each of the possible solutions 36 to the problem. The Figures also illustrate a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform at least the method described above. There is thus described an apparatus 10, method, and computer program with a number of advantages as described above and below. Furthermore, it is significantly more cost effective and less time consuming for virtual consumers according to apparatus 10 to predict the sentiment of real-world consumers towards new services or products rather than to engage directly with real-world consumers. Furthermore, in relation to media and entertainment content, the degree of insight provided by apparatus 10 would be impossible without audience pre-screenings, which are costly, complex, time consuming to administer and possible only once the content is mature. Apparatus 10 provide content creators an accessible and practical means to gain audience insights, iteratively, during the developmental stages of their work. Examples of the disclosure are sufficiently fast and inexpensive to be utilised by all parts of the media and entertainment across the development process for new content. The term ‘comprise’ is used in this document with an inclusive not an exclusive meaning. That is any reference to X comprising Y indicates that X may comprise only one Y or may comprise more than one Y. If it is intended to use ‘comprise’ with an exclusive meaning, then it will be made clear in the context by referring to “comprising only one..” or by using “consisting”. In this description, reference has been made to various examples. The description of features or functions in relation to an example indicates that those features or functions are present in that example. The use of the term ‘example’ or ‘for example’ or ‘can’ or ‘may’ in the text denotes, whether explicitly stated or not, that such features or functions are present in at least the described example, whether described as an example or not, and that they can be, but are not necessarily, present in some of or all other examples. Thus ‘example’, ‘for example’, ‘can’ or ‘may’ refers to a particular instance in a class of examples. A property of the instance can be a property of only that instance or a property of the class or a property of a sub-class of the class that includes some but not all of the instances in the class. It is therefore implicitly disclosed that a feature described with reference to one example but not with reference to another example, can where possible be used in that other example as part of a working combination but does not necessarily have to be used in that other example. Although examples have been described in the preceding paragraphs with reference to various examples, it should be appreciated that modifications to the examples given can be made without departing from the scope of the claims. For example, in addition to the sets of unidimensional problems 34 / possible solutions 36 data gathered and processed as described above, multidimensional problem 34 / possible solutions 36 consumer data can be gathered, for those groups of interest. This data can be similarly gathered through observation and / or survey, presenting consumers from a group of interest with fully prototyped or completed offerings to gather their real-world distribution of reactions. Features described in the preceding description may be used in combinations other than the combinations explicitly described above. Although functions have been described with reference to certain features, those functions may be performable by other features whether described or not. Although features have been described with reference to certain examples, those features may also be present in other examples whether described or not. The term ‘a’ or ‘the’ is used in this document with an inclusive not an exclusive meaning. That is any reference to X comprising a / the Y indicates that X may comprise only one Y or may comprise more than one Y unless the context clearly indicates the contrary. If it is intended to use ‘a’ or ‘the’ with an exclusive meaning, then it will be made clear in the context. In some circumstances the use of ‘at least one’ or ‘one or more’ may be used to emphasis an inclusive meaning but the absence of these terms should not be taken to infer any exclusive meaning. The presence of a feature (or combination of features) in a claim is a reference to that feature or (combination of features) itself and also to features that achieve substantially the same technical effect (equivalent features). The equivalent features include, for example, features that are variants and achieve substantially the same result in substantially the same way. The equivalent features include, for example, features that perform substantially the same function, in substantially the same way to achieve substantially the same result. In this description, reference has been made to various examples using adjectives or adjectival phrases to describe characteristics of the examples. Such a description of a characteristic in relation to an example indicates that the characteristic is present in some examples exactly as described and is present in other examples substantially as described. Whilst endeavoring in the foregoing specification to draw attention to those features believed to be of importance it should be understood that the Applicant may seek protection via the claims in respect of any patentable feature or combination of features hereinbefore referred to and / or shown in the drawings whether or not emphasis has been placed thereon. I / we claim:

Images