Skills-based classification of products

Abstract

A method implemented on an exchange platform includes receiving a request to generate an overall build score for a first item. A set of listing data descriptive of the first item is extracted from a listing of the first item. Scores are generated for each of a plurality of build attributes for the first item by a generative machine learning model, wherein each build attribute represents a skill or a technique for building an item. A scoring model determines an overall build score for the first item, based on the plurality of scores for the plurality of build attributes. Based on the overall build score, the exchange platform performs one or more actions with respect to the listing of the first item.

Description

PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1SKILLS-BASED CLASSIFICATION OF PRODUCTSBACKGROUND

[0001] Electronic platforms and marketplaces provide users with a wide variety of items for viewing, sale, or use. Identifying a particular item to acquire on such platforms can be timeconsuming and inefficient. Electronic platforms including listings of items which are populated with information about the item supplied by the lister of the item. Certain types of electronic platforms cater to niche items or categories which may not be provided on a large scale in the marketplace. Users of electronic platforms also may need to take into account the reliability of an individual lister with very little information about how accurate the lister’s description of an item actually is. Verifying the accuracy of the listing and having a means to compare the skill used to fabricate different items is helpful to a user.SUMMARY

[0002] This specification describes machine learning-based techniques for evaluating a level craftsmanship of making or building an item listed on an electronic platform. A generative machine learning model (e g., a large language model (LLM)) is used to identify build attributes used in the fabrication of an item and to provide a score for each of the build attributes for the particular item. The LLM can also be used to generate an overall build score based on all the individual scores for the different build attributes. The methods and systems enable large scaling up of such evaluation for a majority of items listed on the electronic platform. It also enables a quick generation of an overall build score for ranking, filtering or comparing the listed products or items. The overall build score and the individual scores enable improved verification of listing descriptions and individual vendors / listers by confirming that the original listing of the item and any information about the item by the lister correlates with these scores. In addition, comparison amongst different categories of items is enabled by the use of the overall build score.

[0003] Decomposing and evaluating each build attribute separately allows for a more objective assessment of craftsmanship than attempting to evaluate the whole product at once, which typically leads to a very subjective outcome. In principle, the techniques and methods described in this disclosure scale to every use case where it is possible to decompose the main aspect at hand into individual components from a fixed set of components. The overall build score correlates not justPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1with the level of craftsmanship necessary to build an item, but also with the overall time and effort that it took the artisan to craft the item.

[0004] A computer-implemented method, performed by one or more computing devices corresponding to an exchange platform, is presented in this disclosure. The method includes receiving, by an item evaluation engine, a first request to generate an overall build score for a first item provided on the exchange platform, wherein the overall build score represents a proficiency for building the first item provided on the exchange platform. The method includes extracting, by the item evaluation engine and from a listing of the first item in a data repository, a set of listing data descriptive of the first item. The method includes submitting, by the item evaluation engine and to a generative machine learning model, a second request to generate scores for each of a plurality of build attributes for the first item, wherein the second request includes the set of listing data descriptive of the first item and wherein each build attribute represents a skill or a technique for building an item. The method includes generating, by the generative machine learning model and based on the second request, scores for each of the plurality of build attributes for the first item. The method includes generating, by a scoring model and based on the generated scores of the plurality of build attributes, the overall build score for the first item. The method includes performing, by the item evaluation engine and based on the overall build score, one or more actions on the exchange platform and with respect to the listing of the first item.

[0005] Prior to receiving the first request to generate an overall build score, the method may generate, by a build attribute generation engine, a dataset comprising the plurality of build attributes. The one or more actions can include modifying a search algorithm to rank the listing of the first item based on the overall build score for the first item, validating the listing of the first item on the exchange platform by comparing the overall build score with the listing. Based on the overall build score not aligning with the listing of the first item, the one or more actions can include modifying the listing or disabling the listing. In implementations, a second item with an overall build score comparable to the overall build score of the first item may be presented to the user device. The listing of the first item can include an image of the first item and text descriptive of the first item. Extracting the set of listing data descriptive of the first item can include extracting, using an image processing model, build attributes from the image of first item. The text descriptive of the first item can include at least one of dimensions of the first item, an origin of the first item,PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1a design of the first item, a description of how the first item was built, or a material used to build the first item. The scoring model can be implemented as part of the generative machine learning model. The generative machine learning model can be a large language model. The overall build score can be determined by summing the generated scores of the plurality of build attributes. A value of each score of the plurality of build attributes can increase in a geometric fashion from a least important build attribute score to a most important build attribute score. In some implementations, if the build attribute is unnecessary to build the first item then the score can be 0; if the build attribute of first item is a low level then the score can be 1; if the build attribute of the first item is a medium level then the score can be 3; and if the build attribute of the first item is a high level then the score can be 10.BRIEF DESCRIPTION OF THE DRAWINGS

[0006] FIG. 1 illustrates the components of the system for classifying items.

[0007] FIG. 2 illustrates the various components and some details of operation of the system of FIG. 1

[0008] FIG. 3 illustrates a process flow for generating an overall build score.

[0009] FIG. 4 illustrates an example listing of an item made with a high skill level.

[0010] FIG. 5 illustrates an example listing of an item made with a low skill level.

[0011] FIG. 6 illustrates networked electronic devices.

[0012] Like reference numbers and designations in the various drawings indicate like elements.DETAILED DESCRIPTION

[0013] This disclosure relates to methods and systems for providing an overall build score which correlates with the level of skill or craftsmanship used to make an item.

[0014] There are several significant advantages over conventional techniques that this disclosure provides. A conventional solution to the problems described herein is to permit an item lister (e.g.,PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1a lister device) to identify the craftsmanship level of the item being listed or to use third-party human curation to validate the craftsmanship level. Although a lister may identify the level of craftsmanship of a provided item, there may not be a way to evaluate the reliability of the lister’s assertion or of the lister itself, who may not be known by the viewer or another party (e.g., a certifier, a trusted third party, the electronic marketplace, etc.). Verifying the honesty or obj ectivity of an item’ s provided description is thus a resource-intensive task because it often requires multiple people to review each listing and the associated product to evaluate the build quality of an item. This issue is further amplified for a platform that might have millions of items and for which such manual evaluation of a listing and the associated items is too resource intensive. It is therefore very difficult to provide enough resources to verify the large number of items (e.g., tens of millions or hundreds of millions of items) available on electronic platforms. By using a generative machine learning model (e.g., a large language model) the entire process of identifying the level of craftsmanship can be automated. The automation enables a cheap and relatively objective identification of the level of craftsmanship for items listed, subject to the inherent biases of the training data set of the generative machine learning model. The automation also enables verification of trusted item listers by confirming the determined craftsmanship level with the lister’s own description. The automation also aids viewers by providing an additional level of confidence that the item being considered meets the desired level of craftsmanship and is not just based on the lister’s own description. In addition, the system can identify and even correct errors in the lister’s own description. In embodiments, the techniques described in this disclosure, can confirm authenticity by processing images of the item and cross-referencing with information provided by the lister. The techniques can include evaluating the processed images and the cross-referenced information using an LLM to generate build scores for various build attributes and an overall build score.

[0015] In addition to the advantages of verification and trust-building noted above, the methods and system described in this disclosure also enable a comparison of the level of craftsmanship across items of very different categories. For example, it has traditionally been difficult to compare the level of craftsmanship for an item of one category (e.g., embroidered clothing or ceramic pottery) with the level of skill to make than an item made from a completely different category (e.g., embossed metal or wood joinery) since the build methods in the two categories are so different. Human-based comparisons would require that one or several human experts were equallyPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1well versed in two different fields and that the experts all evaluated the two items equivalently. Such dual-area experts are exceedingly rare and too expensive for frequent use. This system enables a standardization of skills across different types of products, which then enables comparison of the skill levels across item categories by comparing the overall build scores of the two different items (regardless of whether they are made using the same techniques or not).

[0016] The techniques and systems described in this disclosure thus provide the advantages of scalability, objectivity, and broader applicability. The techniques and systems of this disclosure enable verification of the objectivity of particular listers by comparing their descriptions of the products they provide with the overall build score of the items they have listed. The overall build score or individual build scores can also be used to enhance search capabilities for a user. In an example, a search algorithm can provide the overall build score as an additional parameter used for searching or filtering of results and can thereby provide more responsive results to user queries.

[0017] FIG. 1 illustrates the components of a system for classifying items based on a listing of the item provided on an exchange platform. FIG. 2 illustrates additional details of the system of FIG.1 using an example item. The system 100 for classification of items includes an exchange platform 110 and can include user devices 140, and lister devices 145 interacting with the exchange platform 110 via a network 190. The user devices 140 can run applications 140-A. The lister devices 145 can run applications 145-A. In some implementations the user device applications 140-A and the lister device applications 145-A include an application provided by the exchange platform 110 to promote ease of interaction. The exchange platform 110 can include one or more computing devices.

[0018] The exchange platform 110 includes a build attribute generation engine 112, a build attribute database 114, an item evaluation engine 116, a generative machine learning model 120, a large language model 122, other model(s) 124, a scoring model 126, a search optimization engine 128, a listing update engine 130, and a listing data repository 132. The build attribute database 114 and listing data repository 132 can reside remotely or locally. The lister device 140 can submit a listing 102 to the exchange platform 110 or to the listing data repository 132 directly through the network 190. For example, the listing 102 can include a description of an item such as a chair, a desk, a table, clothing, etc. The listing 102 can include details of the item including a textPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1description, an image (or multiple images) of the item, a video, an audio file, and the like. The user device 140 can request that the listing 102 be displayed on the user device 140. The user device 140 or the exchange platform 110 can also send a request 104 to generate an overall build score 230 for the item to the exchange platform 110.

[0019] The build attribute generation engine 112 generates a list of build attributes 210. The build attribute generation engine 112 generates a dataset including a plurality of build attributes. Each build attribute represents a skill or a technique for building an item. The list of build attributes 210 can be generalized for building many different types of items. An example list of build attributes 210 can include joinery, sanding, planning, marking, sawing, and other skills related to carpentry. The list of build attributes 210 can further include welding, soldering, brazing, laser cutting, and other skills related to metal work. The list of build attributes 210 can further include embroidering, sewing, knitting, knotting, crocheting, needlework, and other techniques related to clothing or fabrics. The list of build attributes 210 encompasses the skills and techniques for fabricating a large number of the items listed on the exchange platform 110. The build attribute generation engine 112 can generate a list of building attributes 210 prior to a listing 102 being transmitted to the exchange platform 110 and prior to a request 104 being submitted to the item evaluation engine 116. In an example, categories of products of various listed items and the skills and building attributes used to produce the various items can be stored in a build attribute database 114. The build attribute database 114 can be periodically updated over the network 190. The build attribute generation engine 112 can generate the list of building attributes 210 by accessing the data from the build attribute database 114. In an example, the build attribute database 114 can be an electronic encyclopedia (e.g., Wikipedia) and the build attribute generation engine 112 can periodically (e.g., daily, weekly, monthly, annually, etc.) access the build attribute database 114 to update the list of building attributes 210. In some implementations, the build attribute database 114 stores the list of building attributes 210 and is periodically updated over the network 190 by the build attribute generation engine 112 accessing online resources (e.g., Wikipedia or proprietary online encyclopedia).

[0020] The item evaluation engine 116 receives the request 104 to generate an overall build score 230 for a first item provided on the exchange platform 110. The overall build score 230 represents a proficiency for building the first item provided on the exchange platform 110 and can be aPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1measure of the overall quality of the item. The item evaluation engine 116 receives the listing 102 and any additional information about the listing from the listing data repository 132. The additional information associated with the listing 102 can, for example, include meta-data about the lister device 145, date and time the listing 102 was first submitted, reviews associated with the lister device 145, and the like. The listing 102 and the additional information can include text, images, video, audio, or other information.

[0021] The item evaluation engine 116 generates listing data 106 for the specific item. The item evaluation engine 116 extracts from a listing data repository 132, a set of listing data 106 descriptive of the first item. Some of the listing data 106 comes directly from the text of the listing 102. Example listing data 106 include the physical dimensions of the item or the provenance of the item. The text descriptive of the first item can include, e.g., dimensions of the first item, an origin of the first item, a design of the first item, a description of how the first item was built, or a material used to build the first item. The item evaluation engine 116 can include an image processing model 118 trained to analyze images of the listing 102 and determine the listing data 106. Example listing data 106 extracted from images, videos, or audio about the item can include the material(s) used to make the item, the shape of the item, the color of the item, an estimated age of the item, how much wear and tear the item has experienced, and other information related to the item. Extracting the set of listing data 106 descriptive of the first item can include extracting, using an image processing model 118, build attributes from the image of first item. In an example, the image processing model 118 can extract features from the image(s) of the listing 102, label or identify the various extracted features, and analyze the features and labels to provide additional descriptive terms. The image processing model 118 can include a large language model. The image processing model 118 can be a machine learning model trained to process input images and to output data about the different build attributes or features associated with the item. For example, when presented with an image of a chair as input, the image processing model 118 can identify the material used (e.g., wood), the type of skill (e.g., joinery, sanding, staining, etc.), the age, etc.

[0022] The item evaluation engine 116 submits to a generative machine learning model 120 a request 104 to generate scores for each of the plurality of build attributes for the first item identified by the item evaluation engine 116. The request 104 can include the set of listing data 106 descriptive of the first item.PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1

[0023] The generative machine learning model 120 receives the listing data 106 and the build attributes list 210. The generative machine learning model 120 generates scores for each of the plurality of build attributes of the first item based on the request 104. The generative machine learning model 120 generates a build attribute score list 220. The generative machine learning model 120 determines a score associated with each skill in the build attributes list 210. This build attribute score list 220 associates a score with each build attribute of the build attributes list 210 in which the score denotes a proficiency or skill level employed to build that particular item. For example, a rocking chair made entirely of wood by hand tools can include the build attributes of measurement, hand-sawing, sanding, joinery, and staining, but may not include any skill in welding. In this example, the generative machine learning model 120 can assign a value of zero for the build attribute of welding, since welding is not used to fabricate a wood rocking chair. The generative machine learning model 120 can assign various values for the build attributes of measurement, hand sawing, joinery, and staining. In some implementations, the generative machine learning model 120 includes a large language model 122 or another model or other models 124. An example other model 124 is a model specifically trained to identify build attributes for a particular category of items (e.g., pillow covers). The generative machine learning model 120 can be provided with a prompt including the list of build attributes 210 or the generative machine learning model 120 can be provided with access to the list of build attributes 210 during execution. The generative machine learning model 120 can receive as a prompt the listing data 106, the listing 102, the request 104, and other additional information generated by the item evaluation engine 116. The prompt can include a request to provide a score for each of the building attributes from the building attribute list 210 for the item from the listing 102. In some implementations, the generative machine learning model 120 includes a large language model (LLM) 122. The LLM 122 has already been trained and so includes some general knowledge. The LLM 122 generates a list of build attribute scores 220 for each of the build attributes in the build attribute list 210. In the example when the item is a wooden rocking chair, the scores for building attributes related to wood-working would be higher than the scores for building attributes related to metal-working, which can be assigned a score of zero, nor received no assigned score at all.

[0024] In the example illustrated in FIG. 2, the build attributes “measurement” and “joinery” are assigned scores of 10 because a custom-made, all-wood rocking chair made by a master carpenter with hand tools requires a high level of skill in measurement and in joinery. In another example,PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1if the item was initially made with a high level of skill in measurement and in joinery, but the item was broken and then repaired at a lower level of skill then the generative machine learning model 120 may assign the build attribute “joinery” at a low level based on the most recent repair rather than the original fabrication. In another example, the image processing model 118 can provide an estimate of the dimensions of the item and can compare this estimate with any dimensions provided as part of the listing 102. In some examples, the dimensions generated by the image processing model 118 can be used to determine the score assigned to the build attribute for measurement for this item. In an example, when the estimated dimensions determined by the image processing model 118 match or are within a predetermined tolerance with the dimensions from the listing 102, then the build attribute for measurement can be assigned a high value. When the estimated dimensions and the dimensions from the listing 102 do not match closely, the build attribute for measurement can be assigned a low value.

[0025] Scoring each build attribute for how important that build attribute is in fabricating the item enables several of the advantages outlined above: verification of quality, verification of lister integrity, and improved search results supplied to the user device. In some implementations, the score can be 0 if the build attribute is not used to fabricate the item. For example, a welding build attribute of joining metal pieces is not used in the fabrication of a wood rocking chair so that the welding build attribute can be assigned a score of 0 for a wood rocking chair. In some implementations the score for each build attribute can be normalized between 0 and 1. In some implementations, a score of zero corresponds to the build attribute not being required to make the item or the item being built with no skill (or very, very low skill) at all. In some implementations, the score of 1 corresponds to an intricately crafted item requiring advanced mastery and much time for many different build attributes. In some implementations the levels of build attribute used to fabricate the item can be subdivided into multiple levels, for example 3 levels, 4 levels, or 10 levels and a score is associated with each build attribute level. For example, in some implementations there are 5 levels of build attribute and the value of the scores varies linearly so that the lowest score can be 0 (not needed to make the item) and the highest score can be 4 (highest build attribute level used to make the item). In some implementations, the value of the scores can vary geometrically or logarithmically. In some implementations, a value of the score increases in a geometric fashion from a least important build attribute score to a most important build attribute score. In an example in which a build attribute has four levels — not used (i.e., the build attributePCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1is unnecessary to build the first item), low level, medium level, high level - the ‘not used’ level can be assigned a score of 0, the low level can be assigned a score of 1, the medium level can be assigned to score of 3, and the high level can be assigned a score of 10. In some implementations there can be four levels and the scores can be 0, 1, 10, and 100 from the least build attribute level to the highest build attribute level. The generative machine learning model 120 produces a score for each build attribute in the build attributes list 210. A build attribute score list 220 of the assigned score associated with each build attribute is shown in FIG. 2.

[0026] In implementations in which the generative machine learning model 120 is a large language model, an example prompt to the generative machine learning model 120 is given below, but other prompts can also be used.Attached is a list of distinct skills involved in crafting products for listing on a website. Given information available in the listing, rate the level of proficiency that is necessary to attain in every skill in the list in order to build the associated product. Use the following scale:• if the skill is not needed to build the product, return 0• if a low level of proficiency in the skill is necessary, return 1• if a medium level of proficiency in the skill is necessary, return 3• if a high level of proficiency in the skill is necessary, return 10.

[0027] The scoring model 126 receives the build attribute score list 220 generated by the generative machine learning model 120. The scoring model 126 determines an overall build score 230 for the item identified in the request 104 based on the generated build attribute score list 220. The scoring model 126 generates the overall build score 230 for the first item based on the generate scores of the plurality of build attributes. In some implementations, the scoring model 126 sums the individual scores for all build attributes to generate the overall build score 230. In some implementations, the overall build score 230 is determined by summing the scores of the plurality of build attributes. In some implementations, the scoring model 126 determines a weighted average of the individual scores as the overall build score 230. The weights for the different build attributes can be determined by the generative machine learning model 120 or by other means. In some implementations, the scoring model 126 is a trained machine learning model and uses machine learning methods to determine the overall build score 230. In some implementations, the scoring model 126 is implemented as part of the generative machine learning model 120. In some implementations, the scoring model 126 is a large language model. In some implementations thePCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1scoring model 126 uses the same large language model as the generative machine learning model 120. The system and method are not limited thereto and the scoring model 126 and the generative machine learning model 120 can be different models including different large language models.

[0028] The item evaluation engine 116, the search optimization engine 128, or the listing update engine 130 performs one or more actions on the exchange platform 110, based on the overall build score 230 and with respect to a listing 102 of the first item. In some implementations, the search optimization engine 128, in response to the request 104 from a user device 140 uses the overall build score 230 to rank items on the exchange platform 110 when generating the response to the request 104. In some implementations, the listing update engine 130 modifies the listing 102 by adding a flag to the listing 102 with information pointing out an inconsistency between the text of the description of the listing 102 provided by the listing device 145 and the overall build score 230 or pointing out an inconsistency between any of the individual build attribute scores 220 and the description of the listing 102. In some implementations, the search optimization engine 128 incorporates the overall build score 230, or individual build attribute scores, in training a search model to respond to a query from a user device 140. In some implementations, a search model ranks the listing 102 of the first item based on the overall build score 230 for the first item. Some example actions can include validating the listing 102 of the first item on the exchange platform 110 by comparing the overall build score 230 with the set of listing data 106 or with the listing 102 or with both. In some implementations, if the overall build score 230 or individual build attribute scores do not align with the listing 102 descriptive of the first item, then the listing optimization engine 130 can modify the listing 102, disable the listing 102, notify the listing device 145 of the lack of alignment, notify the user device 140 of the lack of alignment, or other actions. In some implementations, in response to a new request from the user device 140, a second item with an overall build score comparable to the overall build score of the first item can be presented to the user device 140. In some implementations, the set of listing data descriptive of the first item can include an image of the first item and text descriptive of the first item.

[0029] FIG. 3 illustrates a process flow for generating an overall build score. The process flow 300 includes the operations of generating a list of build attributes (310), receiving a request to generate an overall build score (320), extracting listing data for an item (330), generating scoresPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1for the item for each of the build attributes (340), generating an overall build score (350), and performing an action based on the overall build score (360).

[0030] The generation of build attributes (operation 310) includes determining which skills or techniques are used to build various items. The build attributes can include skills and techniques used from a wide range including, for example, metal-working, carpentry, sewing, pottery, and others. The list of build attributes is not limited and can easily be expanded or periodically updated. For example, if low gravity or zero gravity fabrication techniques become popular, then such techniques can be added to the build attributes list. The generation of build attributes can occur prior to any particular listing being added to the electronic platform and prior to receiving a first request to generate the overall build score for a first item. The generation of build attributes can occur by the build attribute generation engine accessing a build attribute database. The build attribute database can periodically access an online encyclopedia (e.g., a proprietary database, a non-proprietary knowledge warehouse) to be updated. In some implementations, the build attribute generation engine can store the build attributes list in the build attribute database. In some implementations, the build attribute generation engine accesses an online encyclopedia and generates an updated list of build attributes, which it then stores in the build attribute database. In some implementations, the build attribute generation engine uses a large language model (LLM) to generate the list of built attributes.

[0031] The receiving of a request (operation 320) for an overall build score includes receiving by an item evaluation engine, a request to generate an overall build score for an item provided on the exchange platform. The request can originate from a user device, from a lister device, from a third party device, or even from the electronic platform itself. The request includes the listing for the item for which the overall build score is requested. The request can also include a type of scoring desired (e.g., weighted average with weights for various build attributes or build attribute categories included). The request can include a request to include the results (e.g., the overall build score or an individual build attribute score) be incorporated into a search query. For example, the user device can submit a request for hand-made wooden rocking chairs, ranked by overall build score, or ranked by advanced joinery score only. The request can include requests for the generation of overall build scores of items from at least two different categories and can includePCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1that the overall build scores be included as fdter (e.g., excluding items with overall build scores below a threshold, or including only items within a range of overall build scores).

[0032] The extraction of listing data for an item (operation 330) includes extracting information from text, image, audio, video, and meta-data of a listing. The extraction of a set of listing data descriptive of the first item can be performed by the item evaluation engine and from a listing of the first item in a data repository. The extraction of listing data can include, for example, applying an image processing model to extract build attributes associated with the item from an image of the item. The extraction of listing data can include, for example, determining at least one of dimensions of the item, an origin of the item, a design of the item, a description of how the item was built, or a material used to build the item based on an image of the item. The extraction of listing data can include meta-data associated with the listing including the lister device associated with the listing, a date and time of the listing, reviews of the listing or the lister device, and the like. In some implementations, the item evaluation engine or an image processing engine receives the images of the listing of the item to generate the listing data. In some implementations, the image processing engine performs segmentation, identifies the various features of the image(s), and analyzes the features to identify, for example, the main material(s) used to fabricate the item, an estimated age of the item, physical dimensions of the item, and the like.

[0033] The generation of scores for the item for each build attribute in the list of build attributes (operation 340) includes determining whether the build attribute was used to build the item and how important that build attribute was in fabricating the item. The generation of scores can include submitting, by the item evaluation engine and to a generative machine learning model, a second request to generate scores for each of a plurality of build attributes for the first item, wherein the second request includes the set of listing data descriptive of the first item and wherein each build attribute represents a skill or a technique for building an item. The generative machine learning model can generate, based on the second request, scores for each of the plurality of build attributes for the first item. The generation of scores includes determining, if the build attribute was used to build the item, how high a proficiency or skill level was used for the build attribute to build the item. In an example, a complicated or advanced joint (e.g., a hand-cut dovetail joint) can be assigned a high score for joinery. A basic joint (e.g., a laser cut box joint) can be assigned a low score for joinery. In some implementations, if a build attribute was not used to build the item, thenPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1the score associated with the build attribute for that item is zero. In some implementations, different skill levels are associated with different scores for each built attribute, based on the listing data for the item. In some implementations, the generation of scores is performed by an LLM of the generative machine learning model. The LLM receives a prompt requesting a score for each of the build attributes of the build attributes list. In addition to the prompt requesting the scores, the LLM also receives as input the list of build attributes and the listing data. The LLM then outputs a score for the level of proficiency for fabricating the item in the listing for each of the build attributes.

[0034] The generation of an overall build score (operation 350) includes accounting for the individual scores for each of the build attributes for the item. The scoring model generates the overall build score for the first item, based on the generated scores of the plurality of build attributes. In an example, a scoring model determines the overall build score by summing the scores for all the build attributes. In an example, the scores may be normalized before they are summed to determine the overall build score. In an example, the scoring model determines the overall build score based on a machine learning model trained to determine the overall build score given input of the entire build attribute score list. In an example, the overall build score can be determined as a weighted average of all the non-zero scores in the build attribute score list. In the case of a weighted average, a set of weights can be assigned to each of the build attributes for a particular item. For example, a request for an overall build score can include, as part of the prompt to the generative ML model, a list of weights for each of the build attributes or a list of weights for various categories of build attributes. In an example, the user device submits a request for an overall build score and includes a list which weights metal-working build attributes as twice the value of other build attributes. In such an example, the request can include how the scoring model functions by multiplying the generated build attribute score list by the appropriate, user device supplied weights before adding all the weighted values together.

[0035] An action performed based on the overall build score (operation 360) can include various options. The item evaluation engine can perform one or more actions on the exchange platform and with respect to the listing of the first item, based on the overall build score. The one or more actions can include modifying a search algorithm to rank the listing of the first item based on the overall build score for the first item; validating the listing of the first item on the exchange platformPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1by comparing the overall build score with the listing; or modifying the listing or disabling the listing based on the overall build score not aligning with the listing of the first item. The electronic platform can, for example, via the listing update engine, modify the listing of the item, temporarily de-list the item, flag the item for further investigation, or flag the lister device for further investigation, based on the overall build score value. In an example, if a lister device is associated with enough items with a large difference between the determined overall build score and the claimed skill level in making the item, then that lister device can be flagged for additional scrutiny or that lister device can automatically have its listings modified to include a warning. A user device can request that listings requested in a query be filtered based on the overall build score or can be filtered by a single score associated with a particular build attribute. The electronic platform can, via the search optimization engine, modify an algorithm for providing search results in response to a query from a user device, to take into account the overall build score associated with each listing. In an example, the electronic platform can provide the filtered list of listings to the user device in which the filter is based on the overall build score for each item. In some implementations, upon receiving a request from a user device, the electronic platform can identify a pair of listings in two very different item categories which each have similar overall build scores. For example, a user device query can be “Please show me a hand-crafted wooden box and a pillow with a needlepoint face and both the box and the pillow have overall build scores equal to or greater than 3.5.” In some implementations individual build attribute scores can be incorporated into a search on the electronic platform so that a skill level for fabrication of multiple items can be matched. In an example “please show me a hand-crafted wooden box with a joinery build attribute score above 2.5 and a needlepoint pillow with a needlepoint build attribute score above 3.7.”

[0036] FIG. 4 illustrates an example listing for a wooden box made with a high skill level. FIG. 5 illustrates an example listing for a wooden box made with a low skill level. Although a single large image 410, 510 is shown, the full data of the listings includes multiple images 420, 520 of each of the items. The multiple images 420, 520 permit including details showing, for example, a closeup of the joint to show the level of detail involved in fabricating each wooden box. The listing for the high skill level wooden box includes text 430 stating that the box was handmade and also the photographs support this statement since they show intricate dovetail joints. For the high skill box example, the item evaluation engine 116 can extract the listing data which includes the inner dimensions of the box (e.g., 9.25 inches by 9.25 inches by 4 inches (D x W x H)), the materialPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1(e.g., cherry wood), and the dovetail joints which require a high level of skill. When added to the prompt to the generative machine learning model, the generated build attribute score list includes, for example, scores of 10 for measurement and joinery and scores of 3 for hand sawing and staining. An overall build score for the high skill box is 72. The listing for the low skill level wooden box includes text 530 stating that the box is laser cut. The images show box joints, which are simpler to make than dovetail joints for the high skill level box. For this low skill box example, the item evaluation engine can extract the listing data which includes the inner dimensions of the item, the outer dimensions of the item, the material (wood), and the type of joint (box joint). When added to the prompt to the generative machine learning model, the generated build attribute score list includes scores of, for example, 1 for measurement and joinery, and 1 for sanding and finishing. An overall build score for the low skill box is 15.

[0037] In this specification the term “engine” is used broadly to refer to a software-based system, subsystem, or process that is programmed to perform one or more specific functions. Generally, an engine will be implemented as one or more software modules or components, installed on one or more computers in one or more locations. In some cases, one or more computers will be dedicated to a particular engine; in other cases, multiple engines can be installed and running on the same computer or computers.

[0038] The subject matter and the actions and operations described in this specification can be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. The subject matter and the actions and operations described in this specification can be implemented as or in one or more computer programs, e.g., one or more modules of computer program instructions, encoded on a computer program carrier, for execution by, or to control the operation of, data processing apparatus. The carrier can be a tangible non-transitory computer storage medium. Alternatively or in addition, the carrier can be an artificially-generated propagated signal, e.g., a machine -generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. The computer storage medium can be or be part of a machine-readable storage device, a machine-readable storagePCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1substrate, a random or serial access memory device, or a combination of one or more of them. A computer storage medium is not a propagated signal.

[0039] FIG. 6 is a block diagram of computing devices 600, 650 that may be used to implement the systems and methods described in this specification, as either a client or as a server or plurality of servers. Computing device 600 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. Computing device 650 is intended to represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, smartwatches, head-worn devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations described and / or claimed in this specification.

[0040] Computing device 600 includes a processor 602, memory 604, a storage device 606, a highspeed interface 608 connecting to memory 604 and high-speed expansion ports 610, and a low-speed interface 612 connecting to low-speed bus 614 and storage device 606. Each of the components 602, 604, 606, 608, 610, and 612, are interconnected using various busses, and may be mounted on a common motherboard or in other manners as appropriate. The processor 602 can process instructions for execution within the computing device 600, including instructions stored in the memory 604 or on the storage device 606 to display graphical information for a GUI on an external input / output device, such as display 616 coupled to high-speed interface 608. In other implementations, multiple processors and / or multiple buses may be used, as appropriate, along with multiple memories and types of memory. Also, multiple computing devices 600 may be connected, with each device providing portions of the necessary operations (e.g., as a server bank, a group of blade servers, or a multi-processor system).

[0041] The memory 604 stores information within the computing device 600. In one implementation, the memory 604 is a computer-readable medium. In one implementation, the memory 604 is a volatile memory unit or units. In another implementation, the memory 604 is a non-volatile memory unit or units.

[0042] The storage device 606 is capable of providing mass storage for the computing device 600. In one implementation, the storage device 606 is a computer-readable medium. In various differentPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1implementations, the storage device 606 may be a floppy disk device, a hard disk device, an optical disk device, or a tape device, a flash memory or other similar solid state memory device, or an array of devices, including devices in a storage area network or other configurations. In one implementation, a computer program product is tangibly embodied in an information carrier. The computer program product contains instructions that, when executed, perform one or more methods, such as those described above. The information carrier is a computer- or machine-readable medium, such as the memory 604, the storage device 606, or memory on processor 602.

[0043] The high-speed controller 608 manages bandwidth-intensive operations for the computing device 600, while the low-speed controller 612 manages lower bandwidth-intensive operations. Such allocation of duties is exemplary only. In one implementation, the high-speed controller 608 is coupled to memory 604, display 616 (e.g., through a graphics processor or accelerator), and to high-speed expansion ports 610, which may accept various expansion cards (not shown). In the implementation, low-speed controller 612 is coupled to storage device 606 and low-speed expansion port 614. The low-speed expansion port, which may include various communication ports (e.g., USB, Bluetooth, Ethernet, wireless Ethernet) may be coupled to one or more input / output devices, such as a keyboard, a pointing device, a scanner, or a networking device such as a switch or router, e.g., through a network adapter.

[0044] The computing device 600 may be implemented in a number of different forms, as shown in the figure. For example, it may be implemented as a standard server 620, or multiple times in a group of such servers. It may also be implemented as part of a rack server system 624. In addition, it may be implemented in a personal computer such as a laptop computer 622. Alternatively, components from computing device 600 may be combined with other components in a mobile device (not shown), such as device 650. Each of such devices may contain one or more of computing device 600, 650, and an entire system may be made up of multiple computing devices 600, 650 communicating with each other.

[0045] Computing device 650 includes a processor 652, memory 664, an input / output device such as a display 654, a communication interface 666, and a transceiver 668, among other components. The device 650 may also be provided with a storage device, such as a microdrive or other device, to provide additional storage. Each of the components 650, 652, 664, 654, 666, and 668, arePCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1interconnected using various buses, and several of the components may be mounted on a common motherboard or in other manners as appropriate.

[0046] The processor 652 can process instructions for execution within the computing device 650, including instructions stored in the memory 664. The processor may also include separate analog and digital processors. The processor may provide, for example, for coordination of the other components of the device 650, such as control of user interfaces, applications run by device 650, and wireless communication by device 650.

[0047] Processor 652 may communicate with a user through control interface 658 and display interface 656 coupled to a display 654. The display 654 may be, for example, a TFT LCD display or an OLED display, or other appropriate display technology. The display interface 656 may comprise appropriate circuitry for driving the display 654 to present graphical and other information to a user. The control interface 658 may receive commands from a user and convert them for submission to the processor 652. In addition, an external interface 662 may be provided in communication with processor 652, so as to enable near area communication of device 650 with other devices. External interface 662 may provide, for example, for wired communication (e.g., via a docking procedure) or for wireless communication (e.g., via Bluetooth or other such technologies).

[0048] The memory 664 stores information within the computing device 650. In one implementation, the memory 664 is a computer-readable medium. In one implementation, the memory 664 is a volatile memory unit or units. In another implementation, the memory 664 is a non-volatile memory unit or units. Expansion memory 674 may also be provided and connected to device 650 through expansion interface 672, which may include, for example, a SIMM card interface. Such expansion memory 674 may provide extra storage space for device 650, or may also store applications or other information for device 650. Specifically, expansion memory 674 may include instructions to carry out or supplement the processes described above and may include secure information also. Thus, for example, expansion memory 674 may be provided as a security module for device 650 and may be programmed with instructions that permit secure use of device 650. In addition, secure applications may be provided via the SIMM cards, along with additional information, such as placing identifying information on the SIMM card in a non-hackable manner.PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1

[0049] The memory may include for example, flash memory and / or MRAM memory, as discussed below. In one implementation, a computer program product is tangibly embodied in an information carrier. The computer program product contains instructions that, when executed, perform one or more methods, such as those described above. The information carrier is a computer- or machine-readable medium, such as the memory 664, expansion memory 674, or memory on processor 652.

[0050] Device 650 may communicate wirelessly through communication interface 666, which may include digital signal processing circuitry where necessary. Communication interface 666 may provide for communications under various modes or protocols, such as GSM voice calls, SMS, EMS, or MMS messaging, CDMA, TDMA, PDC, WCDMA, CDMA2000, or GPRS, among others. Such communication may occur, for example, through radio-frequency transceiver 668. In addition, short-range communication may occur, such as using a Bluetooth, WiFi, or other such transceiver (not shown). In addition, GPS receiver module 670 may provide additional wireless data to device 650, which may be used as appropriate by applications running on device 650.

[0051] Device 650 may also communicate audibly using audio codec 660, which may receive spoken information from a user and convert it to usable digital information. Audio codec 660 may likewise generate audible sound for a user, such as through a speaker, e.g., in a handset of device 650. Such sound may include sound from voice telephone calls, may include recorded sound (e.g., voice messages, music files, etc.) and may also include sound generated by applications operating on device 650.

[0052] The computing device 650 may be implemented in a number of different forms, as shown in the figure. For example, it may be implemented as a cellular telephone 680, e.g., a smartphone. In some instances, the computing device 650 may be implemented as a tablet 682. Other types of the computing device 650 can include an extended reality device, e.g., an augmented reality device or a virtual reality device, a personal digital assistant, or another similar mobile device.

[0053] Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, integrated circuitry, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmablePCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1processor, which may be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0054] These computer programs (also known as programs, software, software applications or code) include machine instructions for a programmable processor, and can be implemented in a high-level procedural and / or object-oriented programming language, and / or in assembly / machine language. As used herein, the terms “machine-readable medium” “computer-readable medium” refers to any computer program product, apparatus and / or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term “machine-readable signal” refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0055] While this specification contains many specific implementation details, these should not be construed as limitations on the scope of what is being claimed, which is defined by the claims themselves, but rather as descriptions of features that can be specific to particular embodiments of particular inventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features can be described above as acting in certain combinations and even initially be claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claim can be directed to a subcombination or variation of a subcombination.

[0056] Similarly, while operations are depicted in the drawings and recited in the claims in a particular order, this by itself should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing can be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring suchPCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0057] Particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions recited in the claims can be performed in a different order and still achieve desirable results. As one example, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some cases, multitasking and parallel processing can be advantageous.

Claims

PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO1CLAIMSWhat is claimed is:

1. A computer-implemented method performed by one or more computing devices corresponding to an exchange platform, the method comprising:receiving, by an item evaluation engine, a first request to generate an overall build score for a first item provided on the exchange platform, wherein the overall build score represents a proficiency for building the first item provided on the exchange platform;extracting, by the item evaluation engine and from a listing of the first item in a data repository, a set of listing data descriptive of the first item;submitting, by the item evaluation engine and to a generative machine learning model, a second request to generate scores for each of a plurality of build attributes for the first item, wherein the second request includes the set of listing data descriptive of the first item and wherein each build attribute represents a skill or a technique for building an item;generating, by the generative machine learning model and based on the second request, scores for each of the plurality of build attributes for the first item; generating, by a scoring model and based on the generated scores of the plurality of build attributes, the overall build score for the first item; andperforming, by the item evaluation engine and based on the overall build score, one or more actions on the exchange platform and with respect to the listing of the first item.

2. The computer-implemented method of claim 1, comprising, prior to receiving the first request to generate an overall build score, generating, by a build attribute generation engine, a dataset comprising the plurality of build attributes.PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO13. The computer-implemented method of claim 1, wherein the one or more actions comprise modifying a search algorithm to rank the listing of the first item based on the overall build score for the first item.

4. The computer-implemented method of claim 1, wherein the one or more actions comprise validating the listing of the first item on the exchange platform by comparing the overall build score with the listing.

5. The computer-implemented method of claim 4, wherein, based on the overall build score not aligning with the listing of the first item, modifying the listing or disabling the listing.

6. The computer-implemented method of claim 1, wherein a second item with an overall build score comparable to the overall build score of the first item is presented to a user device.

7. The computer-implemented method of claim 1, wherein the listing of the first item comprises an image of the first item and text descriptive of the first item.

8. The computer-implemented method of claim 7, wherein extracting the set of listing data descriptive of the first item comprises extracting, using an image processing model, build attributes from the image of first item.

9. The computer-implemented method of claim 7, wherein the text descriptive of the first item comprises at least one of dimensions of the first item, an origin of the first item, a design of the first item, a description of how the first item was built, or a material used to build the first item.

10. The computer-implemented method of claim 1, wherein the scoring model is implemented as part of the generative machine learning model.

11. The computer-implemented method of claim 1, wherein the generative machine learning model is a large language model.PCT / US25 / 14909 07 February 2025 (07.02.2025)Attorney Docket No. 47166-0069WO112. The computer-implemented method of claim 1, wherein the overall build score is determined by summing the generated scores of the plurality of build attributes.

13. The computer-implemented method of claim 12, wherein a value of each score of the plurality of build attributes increases in a geometric fashion from a least important build attribute score to a most important build attribute score.

14. The computer-implemented method of claim 13, whereinif the build attribute is unnecessary to build the first item then the score is 0;if the build attribute of first item is a low level then the score is 1;if the build attribute of the first item is a medium level then the score is 3; and if the build attribute of the first item is a high level then the score is 10.

15. A system for performing the computer-implemented method of any of claims 1-14.

Images