Methods and systems for facilitating the creation of product concepts for food and / or beverages.

Abstract

A computer-aided method for facilitating the creation of product concepts for food and / or beverages, comprising the steps of providing a plurality of content items obtained from one or more sources, each content item containing structured or unstructured text, and at least one of the content items containing text at least partially relating to one or more food and / or beverage products; and assigning to each content item one or more classification labels, each label containing a term belonging to a predefined classification of terms, each label assigned to each content item being selected based on specific wording used in each content item. A computer implementation method comprising: assigning labels; receiving a query from a user specifying a first term; identifying one or more term pairs, where each term pair includes the first term and one other of the classification labels, if the first term includes a classification label from a predefined term classification; determining a relevance score for each term pair, which is at least in part based on the number of content items to which both classification labels in the pair are assigned; and outputting a response in a user interface that provides a display of the relevance score(s) for each term pair in response to the query.

Description

Technical Field

[0001] Embodiments described in the present specification relate to a method and system for analyzing content items and leveraging this analysis to develop or refresh consumer product concepts, in particular food or beverage product concepts. Background Art

[0002] Social listening is widely used to understand the performance of a company or brand in the market, or to identify trends, needs, or gaps related to specific topics. For example, by reviewing content items such as content items posted on social media websites, it is possible to grasp people's thoughts and opinions on a wide range of topics or products, including various food and beverage products. The knowledge acquired in this way can be utilized to develop food and beverage products that better meet people's needs and are more likely to be enjoyed by them.

[0003] To date, the analysis of social listening platforms has relied heavily on the manual interpretation of website content, including people's posts and comments. While such analysis can be used to visualize broad trends in sentiment towards various food and / or beverage products, people's preferences are typically influenced by two or more factors (age, geography, etc.). Identifying the connections between these underlying factors can be difficult when relying solely on manual evaluation. The development of artificial intelligence (AI) systems offers new opportunities to identify such connections at a deeper level, but such systems struggle to formulate results in a way that product or concept developers can easily understand and interact with when developing / designing new product concepts or refreshing existing ones. Such systems have also failed to integrate the various product concept dimensions (e.g., raw materials, product form, sensory characteristics, proposed health benefits, etc.) necessary to facilitate the generation of new product concepts for food and beverages. [Overview of the project]

[0004] According to a first aspect of the present invention, a computer-aided method is provided for facilitating the creation of product concepts for food and / or beverages, the method being A process of providing multiple content items obtained from one or more sources, each content item containing structured or unstructured text, and at least one of the content items containing text at least partially relating to one or more food and / or beverage products, The process of assigning one or more classification labels to each content item, each label containing a term included in a predefined classification of terms, and each label assigned to each content item being selected based on the specific language used in that content item, The process involves receiving a query from the user specifying a first term, If the first term includes a classification label from a predefined term classification, The process involves identifying one or more term pairs, each term pair comprising a first term and one other classification label, For each term pair, a relevance score is determined, which is at least partially based on the number of content items to which both classification labels within the pair are assigned. The user interface includes the step of outputting a response that provides a display of relevance scores (multiple) for each term pair, which is a response to a query.

[0005] If the first term in the query does not contain one of the classification labels in a predefined classification, this method will The process of identifying one or more content items to which the first term relates, based on the specific wording used in each content item, The process involves identifying one or more term pairs, each term pair comprising a first term and one of the classification labels, For each term pair, the process involves determining a relevance score, which is at least partially based on the number of content items to which the first term is identified as relevant and to which the classification label included in the pair is assigned. The user interface may include a step of outputting a response that provides a relevance score(s) for each term pair, which is a response to a query.

[0006] This method may include, for each classification label, the step of identifying the broadest class of terms in a predefined classification to which the classification label belongs. When outputting responses to queries, the classification labels for each term pair are grouped based on the class of the larger label to which the classification label belongs.

[0007] A predefined classification may include multiple classification structures, each structure including a hierarchy of terms, and the terms in each classification structure are: raw materials, consumption scene, Mental or physical needs, Product form, packaging form, Sensory characteristics, and Related to one or more of the following: geographical location.

[0008] In a preferred embodiment, the predefined classification may include a plurality of classification structures, each structure including a hierarchy of terms, and the terms in each classification structure are: Ingredients and Consumption scenarios and Related to mental or physical needs.

[0009] In a further preferred embodiment, the predefined classification may include a plurality of classification structures, each structure including a hierarchy of terms, and the terms in each classification structure are: Ingredients and Consumption scenarios and Mental or physical needs, Product form and related to

[0010] In a further preferred embodiment, the predefined classification may include a plurality of classification structures, each structure including a hierarchy of terms, and the terms in each classification structure are: Ingredients and Consumption scenarios and Mental or physical needs, Product form and, Related to packaging form.

[0011] In a further preferred embodiment, the predefined classification may include a plurality of classification structures, each structure including a hierarchy of terms, and the terms in each classification structure are: Ingredients and Consumption scenarios and Mental or physical needs, Product form and, a packaging form, a sensory characteristic, and a geographic location.

[0012] The response may include, for each classification label included in a respective one of the term pairs, both a display of the classification label and a relevance score between the label and the first term.

[0013] The method comprises: constructing a knowledge graph, the knowledge graph comprising two or more nodes and one or more edges, wherein each node is associated with a respective label in a predefined term classification, each edge connects a respective pair of nodes whose labels are both applied to the same one of the content items, and each edge indicates a relevance score of the labels associated with the nodes in the pair. Outputting the response to the query comprises identifying a first node associated with a first label and one or more other nodes connected to the first node by edges, and determining the relevance score of the edges connecting the first node with the one or more other nodes.

[0014] The method may comprise: receiving, from a user, a selection of one of the pairs of labels that has a relevance score; and retrieving one or more of the content items to which both labels in the pair of labels are assigned, for display to the user.

[0015] The method may comprise: receiving a further plurality of content items over time; assigning one or more of the labels in the classification to each of the further plurality of content items; and updating the one or more relevance scores based on the labels assigned to the further plurality of content items.

[0016] The method is, The process may include outputting a display of the change in relevance score(s) that occurred over a defined period for each term pair.

[0017] Multiple content items, social media sites, News website, blog, Review website, Video or image website, Scientific publications website or repository, web forum, Documents available internally, and It can be obtained from one or more bulletin board systems (BBS).

[0018] The method may include generating product concepts, particularly food and / or beverage product concepts, the product concepts including a description of a food and / or beverage product incorporating one or more ingredients associated with the label output in the response.

[0019] More specifically, a product concept, particularly a product concept for food and / or beverages, includes a description of the food and / or beverage product incorporating one or more ingredients associated with the label output in the response, and this description is structured to include paragraphs describing “Insight,” paragraphs providing “Product Description,” paragraphs describing “Benefits,” and paragraphs describing “Reasons to Believe.” Figure 9 shows an example of such a description that forms a product concept.

[0020] In some embodiments, the paragraphs corresponding to “Product Description” and “Benefits” may be part of the same paragraph. The different paragraphs mentioned above (i.e., the paragraphs related to “Insights,” “Product Description,” “Benefits,” and “Why Believe It Is”) may be separate or combined into a single text.

[0021] "Insights" define the context in which the product is being developed (essentially, the "problem" that the developers hope to solve with the new product). "Product Description" and "Benefits" define the nature of the product, particularly in terms of structural, sensory, emotional, and beneficial features. "Reasons to Believe" provides justification for why the product is likely to meet consumer expectations and successfully solve problems, including in specific markets.

[0022] Product concepts, particularly those for food and / or beverages, can be generated at least partially using a large-scale language model (LLM), and the labels output in the response are used as input to the LLM.

[0023] A second aspect of the present invention provides a computer system comprising one or more processors configured to perform the method according to the first aspect of the present invention.

[0024] A third aspect of the present invention is provided, which stores computer-executable instructions that, when executed by a computer, cause the computer to execute the method according to the first aspect of the present invention. [Brief explanation of the drawing]

[0025] Next, embodiments of the present invention will be described as examples with reference to the attached drawings. [Figure 1] This figure shows a schematic diagram of a system according to one embodiment. [Figure 2] This figure shows a series of steps performed by the system of claim 1 according to one embodiment. [Figure 3] This figure shows an exemplary classification used to label content items in one embodiment. [Figure 4A] This figure shows an example of content items labeled using classification labels according to one embodiment. [Figure 4B] This figure shows an example of content items labeled using classification labels according to one embodiment. [Figure 4C] This figure shows an example of content items labeled using classification labels according to one embodiment. [Figure 5] This figure shows an example of a knowledge graph generated using the content items in Figures 4A to 4C. [Figure 6] Figure 5 shows an example of how the relevance score between nodes in the knowledge graph can be calculated. [Figure 7] This figure shows the information displayed in the user interface, obtained by exploring the knowledge graph in Figure 5. [Figure 8] This figure shows the information displayed in the user interface, obtained by further exploring the knowledge graph in Figure 5. [Figure 9] This figure shows an exemplary product concept that may be obtained using the embodiments described herein. [Modes for carrying out the invention]

[0026] Where used herein, terms such as “comprise,” “comprising,” etc., should be interpreted not as exclusive or exhaustive, but as encompassing, i.e., “including, but not limited to.” Similarly, terms such as “include,” “including,” and “or” should all be interpreted as encompassing, unless such interpretation is clearly prevented by the context. However, the systems, products, and methods disclosed herein may lack any elements not specifically disclosed herein. Therefore, disclosures of embodiments using the term “comprising” include disclosures of embodiments that “consist essentially of” the specified components, and embodiments that “consist of” the specified components.

[0027] As used herein, the singular articles "a," "an," and "the" include multiple subjects unless the context explicitly states otherwise. Thus, for example, a reference to "(a) term" or "(the) term" includes one term, but also two or more terms. Unless otherwise defined, all technical and scientific terms have the same meaning as commonly understood by those skilled in the art in the field to which this invention belongs, and should be given the same meaning.

[0028] Where used herein, the term "and / or" in the context of "X and / or Y" should be interpreted as "X" or "Y," or "X and Y." Similarly, "at least one of X or Y" should be interpreted as "X" or "Y," or "both X and Y." For example, "food and / or beverage" means "food" or "beverage," or "both food and beverage."

[0029] Where used herein, the terms “example,” “etc.,” and “for example,” particularly when followed by a list of terms, are merely illustrative and descriptive, and should not be considered exclusive or inclusive. However, disclosures of embodiments using the terms “example” and “for example, etc.” include disclosures of embodiments in which these terms are exclusive and / or inclusive.

[0030] As used herein, the singular “one” (“a,” “an,” and “the”) includes multiple references unless otherwise indicated. Thus, for example, a reference to “(a) node” or “(the) node” includes one node, but also includes two or more nodes.

[0031] As used herein, the terms “beverage product” and “drink” refer to a product or composition intended for consumption by an individual, such as a human or a non-human animal (e.g., a pet), that provides the individual with at least one nutrient, and that is generally consumed by drinking.

[0032] As used herein, the term “food product” refers to any product or composition intended for consumption by an individual, such as a human or non-human animal (e.g., a pet), and which provides the individual with at least one nutrient, distinct from “beverage” and “drink.” To avoid ambiguity, the term “food product” also includes pet food and nutritional products. Nutritional products may be selected from the list of foods, or dietary supplements, nutritional compositions, complete nutritional compositions, pharmaceuticals, oral nutritional supplements, medical foods, nutraceuticals, powdered nutritional products that are reconstituted with water or milk before ingestion, infant formula, foods for special medical purposes (FSMP), medicinal products, infant formula, baby food, and follow-on formulas.

[0033] As used herein, the term “raw materials” means compounds that may be used in the formulation of a product, in particular a food or beverage product, such as biological, mineralogical or chemical compounds, or combinations of such compounds (e.g., mixtures, blends, semi-finished products, etc.).

[0034] As used herein, the term “Consumption Occasions” refers to the events, moments, or situations in which a product, particularly a food or beverage product, is consumed. An example of a consumption occasion is breakfast.

[0035] As used herein, the term “mental or physical needs” refers to needs relating to the mental and physical elements, states and / or conditions of an object, particularly a human or a non-human animal such as a pet. For example, mental or physical needs may relate to the maintenance, improvement, reduction, correction, removal, addition, replacement, cure and / or prevention of the mental and physical elements, states and / or conditions of an object, particularly a human or a non-human animal such as a pet. For example, physical needs may be immune enhancement, fatigue reduction, or weight maintenance. For example, mental needs may be mood improvement or indulgence enhancement.

[0036] As used herein, the terms “mental or physical needs” and “mental / physical needs” are interchangeable.

[0037] As used herein, the term “product form” refers to the form of a product, particularly a food or beverage product (e.g., beverage, powder, scoopable, etc.), type (e.g., fermented, ...), and storage conditions (e.g., room temperature, frozen, refrigerated, etc.).

[0038] As used herein, the term “packaging form” refers to information relating to the packaging in which a product, in particular a food or beverage product, is supplied (e.g., bottle, split pot, etc.) or alternatives to other packaging forms (e.g., bulk), such as material, shape, packaging type, etc.

[0039] As used herein, the term “Sensory” refers to information relating to the texture, taste, flavor, smell, sensation, aroma, and appearance of a product, particularly a food or beverage product.

[0040] As used herein, the term “geographic location” refers to any type of geographic location, regardless of its size. In non-limiting examples, it could refer to a continent (e.g., the European continent), a geographical region (e.g., South America), a country (e.g., France), a city (e.g., Paris, Châteauneuf-du-Rêve), a place (e.g., a home, a cinema, a school), a geographical structure (e.g., an island, a mountain), or a geographical division (e.g., French-speaking countries).

[0041] Figure 1 shows an overview of System 100 according to one embodiment. The system comprises a pipeline of components that can collectively support product or concept developers in generating new concepts for product development or product innovation by providing relevant data-driven insights. Given a partially defined concept scope (market, topic, raw materials, credible reasons, mental / physical needs, persona, product form, product category, consumer concerns, and benefits), the system can be used to help product or concept developers refine that scope until the product concept is fully developed. As shown in Figure 1, the system comprises a content analyzer 101, a classifier generator 103, a knowledge graph generator 105, and a user interface 107.

[0042] The role of each component in Figure 1 can be further understood by referring to Figure 2, which shows a series of steps performed by different components. In step S201, the content analyzer receives input data in the form of content items. Content items may be obtained from multiple sources, for example, content published on social media, news websites, blogs, forums, scientific publications, and bulletin board systems (BBS), as well as other internally available documents (i.e., online or non-online documents, preferably non-online documents available to only a limited number of people). Content items may be provided in either structured or unstructured text format and may be available from one or more information databases. Content may also originate from non-text sources such as audio, video, or images. However, in the case of non-text sources, a text transcript of the non-text source, such as a speech transcript, is provided as a content item in the computer implementation. In a preferred embodiment, media content items include online media content items. More preferably, online media content items include one or more online media content items selected from a list consisting of messages and questions posted on social media websites, content and comments published on blogs, messages and questions posted on forums, text from scientific publication databases, content from bulletin board systems, and combinations thereof.

[0043] In step S202, a preprocessing step is performed in which content items are classified by topic through a high-level analysis of the wording contained in the content items. For example, content items may be processed to separate and discard those that are considered irrelevant to food and / or beverages. The remaining topics may be further classified into different subtopics, such as vegan / vegetarian or infant nutrition. Step S202 is an optional step, and it will be understood that in some embodiments, this initial preprocessing stage may be omitted depending on the nature and scope of the input content items received by the content analyzer 101.

[0044] In step S203, the content analyzer 101 further processes each content item using one or more natural language processing (NLP) algorithms, such as Bert embedding, UMAP, and HDBSCAN. The NLP algorithm(s) may use a combination of syntactic and semantic analysis to identify the words and terms that appear in each content item and extract the overall meaning of those terms in the context of the content item. The NLP algorithm(s) may also analyze the relationships between different words. For example, a content item that states, "Chocolate may help improve your mind-set," may be interpreted as meaning that "chocolate" has a positive relationship with "mind-set."

[0045] After performing semantic analysis on the input media content items, the content analyzer uses classifications provided by the classifier to label the various terms within each media content item. As will be explained in more detail below, the classifications include a hierarchy of terms, which allows the system to identify different terms used throughout the media content item as belonging to a class of one or more elements.

[0046] In step S204, after tagging content items with classification labels, the labeled posts are used by the knowledge graph generator 105 to compile the knowledge graph 105. The knowledge graph uses the information contained in individual content items, along with the classification labels applied to each content item, to form relationships between different terms, and at the same time, distinguish terms that are more strongly related to each other from terms that have weaker or more superficial connections. The knowledge graph essentially provides a repository of information that can be explored by product or concept developers to understand which elements and concepts are more closely related to each other and should be taken into consideration when developing / designing product concepts in relation to a particular topic, or when refreshing existing product concepts (step S205). As described above, the knowledge graph is built on a large amount of data, which may include both public and private content, including information on consumer needs, preferences, and pain points. The knowledge graph continues to evolve over time. Because the underlying data of a knowledge graph includes data reflecting people's psychology, preferences, emotions, and spontaneous opinions on a particular topic, a knowledge graph provides the relationship(s) of key elements / concepts that cannot be easily identified using traditional methods based solely on human evaluation. These relationship(s) of elements and concepts can then be taken into consideration when developing / designing product concepts related to a particular topic, or when refreshing existing product concepts.

[0047] Figure 3 shows an exemplary classification provided by the classification generator 103. The classification itself is formed from a series of different classification structures, each structure defining a hierarchy of terms associated with a particular dimension of the product concept. Such dimensions may include, for example, “raw materials,” “consumption context,” and “mental / physical needs.” As an example, referring to Figure 3, the classification structure for “raw materials” includes the terms “papaya” and “guava.” These terms belong to the subclass of “tropical fruits,” which itself, along with “vegetables” and “citrus fruits,” is included in the broadest class, “raw materials.”

[0048] It will be understood that individual classification structures may be combined within a single structure, or they may be stored as completely separate structures. For example, referring to Figure 3, the three classification structures shown each have “Consumption Context,” “Raw Materials,” and “Mental / Physical Needs” as their broadest classes, respectively. Each of these classification structures may be stored and / or accessed separately from one another. Alternatively, the three classification structures shown in Figure 3 may constitute part of a single classification database.

[0049] Figures 4A–4C illustrate examples of how individual content items (in this case, posts on one or more social media websites) can be labeled using classifications. Items 401a–431a each represent specific posts collected from social media websites, but it should be understood that these items could be obtained from hosts of other online sites, including blogs, web forums, and news websites. In each case, the content analyzer can recognize the presence of specific words or phrases within the post and tag those words with labels from one or more classification structures, such as those shown in Figure 3. By doing so, the content analyzer can generate multiple label groups 401b–431b, each label group recording a list of classification labels that apply to each of the content items 401a–431a. For example, in the case of the first post 401a, the content analyzer 101 can determine that the term "chocolate" falls within the classification structure "raw materials," while "birthday" falls within the classification structure "consumption context," and "weight management" falls within the classification structure "mental / physical needs." Once compiled, label groups 401b–431b can be stored in memory. These label groups can then be used to analyze the relationships between multiple terms of interest, as will be explained in more detail below.

[0050] For the sake of clarity, Figures 4A–4C show each labeled term as having only a single classification label. For example, the term "papaya" is labeled with the classification label "raw material," while "birthday" is labeled with the classification label "consumption context." In some embodiments, the classification label applied to a term always includes the broadest class within the classification structure to which the term belongs. For example, in the case of "papaya," the broadest class in the classification structure is "raw material," while in the case of "birthday," the broadest class in the classification structure is "consumption context." However, once a term has been identified as belonging to a particular one in the classification structure, it will be understood that the classification label applied to that term may include some or all of the hierarchy of classes to which the term belongs. For example, the term "papaya" is included in the class "tropical fruit," which itself is a subclass of "raw material." Therefore, in some embodiments, the label for the term "papaya" may include both "tropical fruit" and "raw material." A content analyzer and / or classifier generator may maintain a list of the number of content items containing each term in a classification.

[0051] As shown in Figures 4A to 4C, the content analyzer may be able to identify a particular term as belonging to a class even if the exact wording used differs from the wording used in the classification structure. For example, post 413a refers to "increased waist size," while post 415a uses the term "slim down." Here, based on the semantic analysis provided by the NLP algorithm used to contextualize the information in the posts, the content analyzer can determine that the descriptions "increased waist size" and "slim down" are related to weight management, and accordingly label the posts with the label "weight management." From this, it can be further understood that a term tagged as belonging to a particular class in classification can have both positive and negative associations with that class. In the above example, "increased waist size" is understood to be a direct result of weight gain. On the other hand, the term "slim down" is also related to weight management in the sense of losing or avoiding weight gain. In both cases, the wording used in the posts, although different, is understood to be related to the "weight management" class in the classification.

[0052] Figure 5 shows an example of how the information contained in elements 401b-431b, compiled from posts 401a-431a, can be used to construct nodes in the knowledge graph. Here, each node in the graph contains a term within one of the classification structures, which can be used to label one or more content items. A node is connected to other nodes that are considered to be associated with that term, based on the information contained in the content item. For example, the node for "ice cream" is connected to the node for "birthday". This reflects the fact that there are one or more posts (e.g., post 407a) that are tagged with the classification label "ice cream" and also tagged with the label "birthday". Similarly, there is a post (e.g., 411a) that is tagged with the classification label "ice cream" and also tagged with the label "vacation". Therefore, the nodes for "ice cream" and "vacation" are also connected by edges in the knowledge graph.

[0053] The strength of the association between each pair of connected nodes varies depending on how often different terms appear together in different posts. Referring to Figures 4A–4C, we can see that two posts, 407a and 425a, are tagged with both the labels "chocolate" and "birthday," while only one post, 407a, is tagged with both "chocolate" and "ice cream." The higher co-occurrence rate of the labels "chocolate" and "birthday" suggests that "chocolate" is more strongly associated with "birthday" than with "ice cream." (Here, Figures 4A–4C are provided as examples; in reality, the sample size of content items would be much larger to allow for the observation of statistically meaningful differences in co-occurrence rates.)

[0054] When generating a knowledge graph, the knowledge graph generator determines a relevance score for each pair of nodes, based at least partially on the number of times the words or terms with those labels appear together within the same content item. Figure 5 shows the relevance score for each edge in the graph. For example, the higher co-occurrence rate of the labels "chocolate" and "birthday" is reflected in a higher relevance score of "4" compared to the relevance score of "2" for the node "chocolate" and "ice cream".

[0055] It will be understood that the relevance score of a pair of nodes in a knowledge graph can be calculated in several different ways. For example, the relevance score can be calculated as the number of content items that contain the classification labels of both nodes in the pair. In some embodiments, the relevance score of each pair of nodes may also reflect how strongly those terms are related to each other within individual content items. For example, a user might post, "I love chocolate ice cream," while another user might post, "My friend's favorite food is chocolate, but my favorite food is ice cream." Both sentences contain the terms "chocolate" and "ice cream," so both contribute to the relevance score of the pair of nodes "chocolate" and "ice cream" in the knowledge graph. However, the second sentence contributes less to the relevance score of these two nodes than the first. This is because the terms "chocolate" and "ice cream" are more weakly related to each other within that sentence compared to the first.

[0056] In some embodiments, the relevance score may take into account not only the number of times the terms appear together, but also whether the terms are understood to have a positive or negative influence on each other. Figure 6 shows an example of relevance scores between the “diabetes” node and other nodes connected to the “diabetes” node (i.e., the “weight management,” “sugar,” and “stress” nodes). Here, “Hit Count” specifies the number of media content items containing the labels of the two nodes within each pair (“diabetes” and “weight management,” “diabetes” and “sugar,” and “diabetes” and “stress”). The hit count can be subdivided into examples of “positive relevance,” “negative relevance,” and “neutral relevance.” If a media content item suggests that one element has a positive influence on the other, the two labels may be recorded as having a “positive” relevance. Conversely, a negative relevance score means the opposite is true. For example, the sentence “Weight gain is bad for diabetes” is recorded as having a negative relevance as a co-occurrence of “weight management” and “diabetes.” The relevance score may be calculated using one of several different functions. In the example shown in Figure 6, the relevance score is obtained by doubling the score for "positive" hits and adding the scores for "neutral" and "negative" hits. This is just an example, and it should be understood that in practice, the final relevance score can be generated in several different ways by biasing both positive and negative relevances depending on the type of trend the user wants to consider.

[0057] Here is an example of how the knowledge graph in Figure 5 can be explored to help develop / design new product concepts or refresh existing ones. Referring again to Figure 1, a product or concept developer (hereinafter, "developer") can interact with the knowledge graph using the user interface 107. To do so, the developer can first log in to an account that stores a list of current and past projects, where each project is related to, for example, a different product concept. The developer can then choose to create a new project flow or continue working on an existing project flow.

[0058] For example, a developer might begin by considering and creating a product concept that could target people with diabetes. Figure 7 shows an example of the results returned from the knowledge graph in Figure 5 when the term "diabetes" is entered as a filter. By entering "diabetes" as a filter, the system allows the user to view the terms in the knowledge graph that are most strongly associated with "diabetes" based on what is stated within the library of content items. For example, the user interface might display the top n terms ranked by the strength of their relevance score to "diabetes," where n is a number specified by the product or concept developer, e.g., 10, 20, 50.

[0059] In the example shown in Figure 7, the results are grouped into three categories: “Ingredients,” “Consumption Situation,” and “Mental / Physical Needs,” which are the same as the conceptual dimensions shown in Figures 4A-4C and the classification structure shown in Figure 3. Essentially, the labels returned from the knowledge graph and displayed along with the correlation score to “Diabetes” (e.g., “Chocolate,” “Birthday,” “Fatigue”) are classified into separate groups based on the broadest class of labels to which each belongs (“Ingredients,” “Consumption Situation,” “Mental / Physical Benefits”).

[0060] Continuing to refer to Figure 7, we can see that the "Ingredients" category returns high results for "Chocolate," low results for "Sugar," and even lower results for "Ice Cream." The size of each bar represents the relevance score between the node for the term "Diabetes" and the respective nodes for the terms "Chocolate," "Sugar," and "Ice Cream." For example, the bar size may be equal to or proportional to the relevance score. The results are depicted as a bar graph in Figure 7, but this is merely an example, and it will be further understood that several different graph formats, including but not limited to pie charts and vertical bar graphs, could be used. Each of these labels is recognized as representing a term within the classification structure of "Ingredients," and is therefore displayed in the "Ingredients" graph. Similarly, the "Consumption Situations" graph shows a strong relevance score between "Diabetes" and "Birthday," while the "Health" graph shows a strong relevance score between "Diabetes" and "Weight Management," and lower relevance scores between "Diabetes" and "Fatigue," and "Diabetes" and "Stress." This shows that queries containing terms from a specific classification structure, such as the "raw materials" classification structure, can yield results from both that same classification structure and other classification structures. For example, a query containing the term "kale" (raw material) could return "lemon" (i.e., another type of raw material), as well as "diabetes" (i.e., a term that falls within a classification structure consisting of terms associated with health and physical benefits).

[0061] Figure 8 shows how product or concept developers can further navigate the knowledge graph by including "weight management" as an additional filter (i.e., by selecting "diabetes" OR "weight management" as the query). By doing so, the "Ingredients" graph will display labels associated with either "weight management" or "diabetes" in the knowledge graph. Thus, in addition to "chocolate," "sugar," and "ice cream," the "Ingredients" graph will include "papaya," "lettuce," "bacon," and "tomato." Similarly, the "Situations" graph will include the labels "vacation," "wedding," and "winter." Since "weight management" is not associated with any labels other than "diabetes" in the "health" class, the "health" graph will not include any further labels compared to Figure 7, although "weight management" is excluded from this graph.

[0062] In some cases, product or concept developers may formulate queries using "AND" statements (e.g., "diabetes" AND "weight management"). In this case, the system needs to identify content items that contain both of these labels and then output any other labels that appear within the same content items. The system can easily do this by referring to the set of label groups 401b-431b. The system first identifies a label group that contains both the label "diabetes" and (AND) the label "weight management" (in the example shown in Figures 4 and 5, this returns label group 421b). The content analyzer can then determine what other terms are present or associated with the content item 421a from which label group 421b is derived. In this way, the content analyzer can identify "chocolate" and "sugar" as output terms for the query because both of these terms appear in at least one content item that has both classification labels, "diabetes" and "weight management". Therefore, by labeling each original content item with a set of classification labels and storing each label group in memory, the system can quickly and easily return results for queries containing two or more terms.

[0063] It will become clearer that queries entered into the user interface by product or concept developers do not need to be one of the classification labels themselves. If a product or concept developer enters a term that is not included in any of the classification structures, the system returns to the original set of input content items, filters those items to identify the items containing that term. The system then determines the classification label that most frequently appears in the label group of the filtered content items. For example, a product or concept developer might enter the term "movie," which is not included in any of the classification structures. In that case, the system filters the original set of content items to identify all content items containing the term "movie," and examines the label groups associated with the filtered set of content items. The system might determine that many of those label groups contain the classification labels "candy," "popcorn," and "Christmas." The system then may display those labels as a result of the product or concept developer's query. The same applies, for example, to countries or geographical locations. That is, a product or concept developer can type in the name of a country or geographical location, and the system filters the content items to determine the label that most frequently appears in relation to that country or geographical location.

[0064] The results obtained in this way in Figure 8 can be useful in helping product or concept developers conceive and develop new product / product concept ideas that align with current consumer needs suggested by various different content streams. For example, a product or concept developer might start by focusing on "diabetes" and thereby identify the association between "diabetes" and "weight management." While it might have been expected that some of the ingredients that appeared when searching for "diabetes" (e.g., "chocolate" and "sugar") would overlap with "weight management," the product or concept developer might be more surprised to see the label "papaya" returned when adding the "weight management" filter. The product or concept developer could use this to further investigate the relationship between "weight management" and "papaya" and view some of the underlying posts that support that relationship, such as those shown by element 415a in Figure 4B and element 429a in Figure 4C. From this, the product or concept developer could understand that people are talking about papaya as being useful for weight management. Given that chocolate is a highly desirable product for consumers, a product or concept developer might use insights gained from a knowledge graph to conceive a concept consisting of a chocolate-based product suitable for people with diabetes and containing papaya extract or pieces of papaya. In a further example, a product or concept developer might focus on the association between "weight management" and "winter" and consider developing a product concept that is particularly appealing during that time of year, such as hot chocolate containing papaya extract.

[0065] To further assist product or concept developers in identifying evolving trends in relevance, knowledge graphs can not only include information about the relevance between different words and phrases, but also maintain records of how relevance scores change over time. In the example above, the knowledge graph may show that the relevance score between "weight management" and "papaya" was value 1 12 months ago, but then surged to a much higher value 2 in the most recent 2-3 months. The changes in relevance scores over time can be displayed in a graph format in the user interface that is easily interpretable by product or concept developers and can be used to support decisions on how to proceed with developing products that embody those characteristics.

[0066] In some embodiments, after using the system to identify specific relationships between terms, product or concept developers can use these to construct a product concept, an example of which is shown in Figure 9. The product concept includes “insights,” which set the background for the product's development (essentially, the “problem” the developer wants to solve with the new product). The concept further includes “product description” and “benefits,” which define the product’s properties, and “credible reasons,” which provide justification for why the product is likely to meet consumer expectations and successfully solve the problem, including a specific market. Figure 9 shows an example where results from Figure 8 were used to identify the relationship between papaya and weight management. This is linked to a desire to provide a chocolate product suitable for people with diabetes. Thus, the results output by the system can be used by product or concept developers to conceive of a hot chocolate product that is suitable for people with diabetes and can help with weight management.

[0067] In some cases, the process of building a product concept based on user interface output can be automated, for example, by using one or more generation algorithms built around a Large-Scale Language Model (LLM). A product or concept developer may select a list of relevant terms provided in the output for their query and input those terms into the LLM, along with instructions to generate one or more paragraphs of text and / or images describing the product concept, which will become the LLM's output. To customize the LLM's output to a desired format and content, the product or concept developer may be provided with options to modify certain parameters of the LLM. For example, the product or concept developer might consider changing the degree of randomness (or temperature) of the LLM, i.e., the degree to which the LLM is constrained by specific terms output from the knowledge graph when generating text. Alternatively, or additionally, the product or concept developer may change the type of prompts used when instructing the LLM to generate text. The product concept can then be iteratively refined by the product or concept developer working in conjunction with the LLM. Examples of suitable LLMs include GPT 3.5, GPT-4, BARD, LLaMA, Falcon, Cohere, PaLM, and Claude V1.

[0068] Accordingly, the embodiments described herein can provide tools for concept development, design, or innovation that enable product or concept developers to gain new insights from available data sources and translate these insights into new product concepts. In doing so, the embodiments can enhance and accelerate the process of generating new concepts or prototypes that resonate with people's current needs in a more data-driven manner.

[0069] Implementations of the subject matter and operations described herein can be realized in digital electronic circuits, or in computer software, firmware, or hardware including the structures disclosed herein and their structural equivalents, or in a combination of one or more thereof. Implementations of the subject matter described herein can be realized using one or more computer programs, i.e., one or more modules of computer program instructions encoded on a computer storage medium for execution by a data processing device or for controlling the operation of a data processing device. Alternatively or additionally, program instructions can be encoded into artificially generated propagating signals, such as mechanically generated electrical signals, optical signals, or electromagnetic signals generated to encode information for transmission to a suitable receiving device for execution by a data processing device. The computer storage medium may be or may be a computer-readable storage device, a computer-readable storage board, a random or serial access memory array or device, or a combination of one or more thereof. Furthermore, although the computer storage medium is not a propagating signal, the computer storage medium may be a source or destination of computer program instructions encoded into artificially generated propagating signals. Computer storage media may also consist of or be comprised of one or more separate physical components or media (e.g., multiple CDs, disks, or other storage devices).

[0070] While specific embodiments have been described, these embodiments are presented only as examples and are not intended to limit the scope of the invention. In fact, the novel methods, devices, and systems described herein can be embodied in a variety of forms. Furthermore, various omissions, substitutions, and modifications in the forms of methods and systems described herein can be made without departing from the spirit of the invention. The appended claims and equivalents are intended to encompass forms or modifications that fall within the scope and spirit of the invention.

Claims

1. A computer-aided method for facilitating the creation of product concepts for food and / or beverages, A step of providing multiple content items obtained from one or more sources, each content item containing structured or unstructured text, and at least one of the content items containing text at least partially relating to one or more food and / or beverage products, The process of assigning one or more classification labels to each content item, each label containing a term included in a predefined classification of terms, and each label assigned to each content item being selected based on specific wording used in each of the said content items, The process involves receiving a query from the user specifying a first term, If the first term includes a classification label from the predefined term classification, The process involves identifying one or more term pairs, each term pair comprising the first term and one other of the classification labels, For each term pair, a relevance score is determined, which is at least in part based on the number of content items to which both classification labels within the term pair are assigned. A user interface includes a step of outputting a response that provides a display of the relevance score (or multiple relevance score) for each term pair, in response to the query. A computer implementation method, including

2. If the first term in the query does not include one of the classification labels in the predefined classification, A step of identifying one or more of the content items to which the first term relates, based on the specific wording used in each of the content items; The process involves identifying one or more term pairs, each term pair comprising the first term and one of the classification labels, For each term pair, a relevance score is determined, which is at least in part based on the number of content items to which the first term is identified as relevant and to which the classification label included in the term pair is assigned. The user interface includes a step of outputting a response that provides a display of the relevance score (or multiple relevance score) for each term pair, The computer implementation method according to claim 1, including the method described in claim 1.

3. For each of the classification labels, the step includes identifying the broadest class of terms in the predefined classification to which the classification label belongs. When outputting the response to the query, the classification label in each term pair is grouped based on the class of the larger label to which the classification label belongs. The computer implementation method according to claim 1 or 2.

4. The aforementioned predefined classification includes multiple classification structures, each structure includes a hierarchy of terms, and the terms in each of the aforementioned classification structures are: raw materials, consumption scene, Mental or physical needs, Product form, packaging form, Sensory characteristics, and geographical location, A computer implementation method according to any one of claims 1 to 3, relating to one or more of the above.

5. The computer implementation method according to any one of claims 1 to 4, wherein the response includes, for each classification label contained in one of the term pairs, both the display of the classification label and the relevance score between the label and the first term.

6. The steps of constructing a knowledge graph comprising two or more nodes and one or more edges, each node being associated with a respective label in the predefined term classification, each edge connecting each pair of nodes where both labels are applied to the same content item, and each edge showing the relevance score of the labels associated with the nodes in the pair, Outputting the response to the query includes identifying the first node associated with the first label, and one or more other nodes connected to the first node by an edge. A step of determining the relationship score of the edge connecting the first node to one or more other nodes, A computer implementation method according to any one of claims 1 to 5, including the method described in any one of claims 1 to 5.

7. The process involves receiving a selection from the user of one of the pairs of labels having a relevance score, The steps include calling one or more of the content items to which both labels in the label pair are assigned in order to display them to the user, A computer implementation method according to any one of claims 1 to 6, further comprising:

8. The process involves receiving multiple additional content items over time, A step of assigning one or more of the labels in the classification to each of the further multiple content items, A step of updating the relevance score(s) based on the labels assigned to the further multiple content items, A computer implementation method according to any one of claims 1 to 7, further comprising:

9. The process of outputting a display of the change in the aforementioned relevance score (multiple scores are possible) over a defined period for each term pair. The computer implementation method according to claim 8, further comprising:

10. The aforementioned multiple content items are, social media sites, News website, blog, Review website, Video or image website, Scientific publications website or repository, web forum, Documents available internally, and Bulletin board system (BBS), A computer implementation method according to any one of claims 1 to 9, obtained from one or more of the following.

11. A computer-aided method according to any one of claims 1 to 10, comprising generating a product concept, in particular a product concept for food and / or beverages, wherein the product concept includes a description of a food and / or beverage product incorporating one or more raw materials associated with the label output in the response.

12. The computer implementation method according to claim 11, wherein the product concept, in particular the product concept for food and / or beverage, is generated at least in part using a large-scale language model (LLM), and the label output in the response is used as input to the LLM.

13. A computer system comprising one or more processors configured to perform the method described in any one of claims 1 to 12.

14. A computer-readable medium that stores computer-executable instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 12.

Images