A multi-source food product element identification method based on dual identity binding and evidence gating

Abstract

The present application relates to food product information processing, multi-source data fusion and entity identity binding technology. The method comprises: establishing a target identity anchor point containing brand identity and SKU identity for the target food product; obtaining supplementary evidence when the basic product fact is insufficient and saving the source and the original text of the evidence; sequentially performing brand identity matching and dish identity matching, and only allowing evidence with double identity matching to enter fact fusion; splitting the evidence matched through identity into atomic facts and performing attribute-level consistency and conflict detection; keeping stable facts supported by multiple parties when conflicts occur, and sending conflicting facts to audit; implementing evidence quality gating according to identity status, source traceability, evidence sufficiency and conflict state; generating traceable SKU element mapping with evidence identification, matching state, conflict state and audit state based on the admitted evidence only. The present application can reduce cross-brand, cross-SKU evidence misbinding and conflict evidence error propagation.

Description

Technical Field

[0001] This invention relates to the fields of natural language processing, information extraction, multi-source data fusion, food product data governance, entity identity binding, and evidence quality control. In particular, it relates to a computer implementation method for performing dual identity matching, atomic fact conflict detection, and evidence gating on menus and external supplementary evidence around the identity anchor point of the target food product, and outputting a traceable food element mapping. Background Technology

[0002] Food product names may be short, marketing-oriented, or have series names or aliases. Multiple SKUs with similar names may also exist under the same brand. Relying solely on dish names for element identification can easily lead to cross-product misidentification.

[0003] When basic menu information is insufficient, supplementary evidence can be obtained from other information sources. However, this supplementary evidence may belong to other brands, another SKU under the same brand, a different product version, or its corresponding target cannot be uniquely identified. Without an identity binding mechanism, external evidence can easily be incorrectly linked to the target SKU.

[0004] Even if the evidence is correctly identified, different sources may still have different values ​​for "included / not included" or the same mutually exclusive attribute. If the system processes the data according to the last obtained result, single source priority, or full-text coverage, conflicting facts may be silently covered and further propagated into the structured food element data.

[0005] Therefore, a processing method is needed that first binds brand identity and SKU identity to evidence before it enters the food element identification process, and then performs atomic fact conflict detection, evidence quality gating and traceability mapping generation after identity confirmation. Summary of the Invention

[0006] The technical problem this invention aims to solve is to reduce the data consistency and traceability issues caused by the misbinding of supplementary evidence across brands or SKUs, the automatic propagation of evidence with uncertain identities, the silent overwriting of conflicting facts from multiple sources, and the lack of an evidence chain in the final element mapping during the processing of multi-source food product information.

[0007] To address the aforementioned issues, this invention provides a method for identifying multi-source food product elements based on dual-identity binding and evidence gating. The method first establishes a target identity anchor for the target SKU; when basic product facts are insufficient to identify the intended food attributes, supplementary evidence is obtained and its source and supporting original text are preserved; subsequently, brand identity matching and dish identity matching are performed sequentially; atomic facts are extracted and conflict detection is performed only on evidence from dual-identity matching; when conflicts occur, stable facts supported by multiple sources are retained and the conflicting facts are entered into review; finally, admission, review, or rejection routing is performed based on the evidence quality status, and a SKU element mapping with an evidence chain is generated.

[0008] By adopting the above scheme, it is possible to prevent the misbinding of evidence across brands and similar SKUs within the same brand before fact fusion; to locate specific attribute conflicts through atomic fact-level conflict detection; to avoid silent overwriting from a single source through stable fact retention and conflict fact diversion; and to improve the consistency and traceability of structured food product element information through evidence quality gating and evidence identification association. Attached Figure Description

[0009] Figure 1 The flowchart below shows the overall process of the multi-source food product element identification method based on dual identity binding and evidence gating provided in this embodiment of the invention.

[0010] Figure 2 This is a schematic diagram of evidence quality gating routing provided in an embodiment of the present invention. Detailed Implementation

[0011] The following, with reference to the accompanying drawings, provides a practically implementable method for illustrating the implementation of brand identity matching, dish identity matching, atomic fact conflict detection, and evidence quality gating. This implementation method is not limited to specific models, similarity algorithms, thresholds, or external search tools; the overall process of this embodiment is as follows: Figure 1 As shown, the evidence quality gating route is as follows: Figure 2 As shown.

[0012] First, establish target identity anchors for each target food product SKU to be processed. Target identity anchors may include the standard brand identifier (canonical_brand_id), brand name (brand_name), brand aliases (brand_aliases), SKU name (sku_name), series name (series_name), official description (official_description), and source context (source_context). Not every field is required, but there should be at least a combination of fields that can identify the target brand and the target SKU.

[0013] For each piece of external supplementary evidence, an evidence record is saved. The evidence record may include source_url, source_type, page_title, detected_brand_name, detected_product_name, detected_series_name, evidence_text, and captured_at. The system retains the original source and supporting evidence text, not just the final conclusions processed by the model or manually.

[0014] In one implementation, the system first performs a sufficiency assessment of the basic product facts. When the dish name, full product name, series name, and product description are sufficient to identify the intended food attributes, element identification can be directly performed based on the basic product facts; when the description is missing, too short, contains only marketing terms, does not match the target dish, or is insufficient to identify the ingredients, flavor, tea base, milk base, fruit, toppings, process, or product form, supplementary evidence acquisition is triggered.

[0015] When matching brand identities, the system first performs character standardization on the brand name, including unifying full-width and half-width characters, uppercase and lowercase letters, and handling spaces and punctuation. It then maps the confirmed official brand abbreviation, Chinese name, English name, and aliases to a unified standard brand identifier. Furthermore, it can remove non-brand identity parts such as city, shopping mall, and store type from the store name.

[0016] When the brand name in the evidence corresponds to the target brand's standard name or a confirmed alias, or when the evidence comes from an official page or application context confirmed to belong to the target brand, and there is no explicit brand identifier that conflicts with the target brand, set `brand_match_status` to `match`. When the evidence clearly points to another brand, set it to `mismatch`. Other cases where unique identification is not possible are set to `uncertain`. Mismatch evidence is rejected, uncertain evidence is submitted for manual review, and only match evidence continues to be used for dish identity matching.

[0017] When matching the identity of a dish, the system standardizes the SKU name and series name respectively, removes spaces and symbols that have no identity differentiation meaning, unifies the confirmed synonyms, retains the core name that can distinguish the product, and saves the series name as an independent field.

[0018] In a deterministic implementation, `dish_match_status` is considered a match if any of the following conditions are met: the standardized full SKU name is identical and there are no conflicts in the series information; the full names are not completely identical, but the core product name and series name are identical; or the core names are identical, the product description in external evidence contains the same identity information as the official description of the target SKU, and there are no conflicts with similar SKUs under the same brand. When evidence clearly corresponds to another SKU under the same brand or when series information clearly conflicts, it is set to mismatch; when it cannot be uniquely confirmed, it is set to uncertain. Mismatch evidence is rejected, and uncertain evidence enters manual review, only allowing fact fusion when both the brand and the dish are match.

[0019] After the dual identity matching is successful, the system breaks down the evidence into atomic facts. An atomic fact can be represented as Atomic_Claim={evidence_id,attribute_type,normalized_value,assertion_type,source_type}. Among them, attribute_type can include objective food attributes such as ingredient, tea_base, milk_base, fruit, topping, flavor, process, and product_form.

[0020] For different valid sources of the same target SKU, the system performs consistency and conflict detection according to attribute level. When one source explicitly claims to contain the target attribute while another source explicitly claims not to contain the target attribute, it is determined to be a conflict; for the same attribute that is predefined as mutually exclusive, if two sources with verified identities give incompatible values, it is determined to be a conflict; if the evidence belongs to different product versions, different SKUs, or the identity cannot be verified, it is transferred to identity or version review, and is not directly determined to be a factual conflict; two atomic facts that can be established simultaneously are not determined to be a conflict.

[0021] The system not only stores conflict identifiers but also the evidence identifiers and atomic facts that caused the conflict, enabling the conflict to be located. When a conflict occurs, instead of forcibly selecting a single source from multiple valid sources to cover the others, it identifies and retains atomic facts that are supported by multiple parties and do not participate in the conflict as stable atomic facts, while conflicting atomic facts are submitted for review.

[0022] For example, three valid pieces of evidence support "oolong tea, osmanthus, fresh milk," "oolong tea, osmanthus," and "oolong tea, osmanthus, plant-based milk," respectively. The system can identify "oolong tea" and "osmanthus" as common stable facts and continue to be considered as candidate facts; it can identify "fresh milk" and "plant-based milk" as conflicting facts and proceed to review, without selecting one source to silently overwrite the other.

[0023] In one implementation, the quality of evidence is set to discrete states. When the source is traceable, the brand matches, the dish matches, and there is objective original evidence supporting the target element without conflict, the quality of evidence is set to high quality; when the identity is basically established but the evidence is insufficient to uniquely support the element's fact, it is set to insufficient evidence; when the identity is established but there is a factual conflict among multiple valid sources, it is set to conflict of evidence; when the brand or dish does not match, the source is untraceable, or no valid evidence can be formed, it is set to rejection.

[0024] Accordingly, high-quality and conflict-free evidence can enter the candidate or approved state; insufficient evidence will be subject to manual review; when sources conflict, stable atomic facts will be retained and conflicting facts will be reviewed; evidence will be rejected when identities do not match. Only atomic facts that pass the evidence gating can form a mapping between the target SKU and the standard element identifier.

[0025] The final traceable mapping can be represented as SKU_Element_Mapping={sku_id,canonical_brand_id,element_norm_id,brand_match_status,dish_match_status,evidence_quality,conflict_flag,manual_review_flag,evidence_ids,adoption_scope,reject_reason,mapping_version}. In actual implementation, field names can be adjusted, as long as the standard element mapping maintains a traceable association with evidence identifiers, identity matching status, conflict status, and review status.

Claims

1. A method for identifying elements in multi-source food products based on dual identity binding and evidence gating, characterized in that, include: Obtain basic product fact information for the target food product SKU and establish a target identity anchor that includes at least both standard brand identity information and target SKU identity information. The system performs an evidence sufficiency assessment on the basic product factual information, triggering supplementary evidence acquisition only when the basic product factual information is insufficient to identify the intended food attributes. For each piece of supplementary evidence, the source information and the original text of supporting evidence are saved. Based on the target identity anchor, brand identity matching is first performed on each piece of supplementary evidence, generating a matching, non-matching, or uncertain brand matching status. Evidence is rejected when the brand does not match, and its automatic propagation is blocked and it enters manual review when the brand is uncertain. Only when the brand matches is the food identity matching process continued. For brand-matched supplementary evidence, food identity matching is performed, generating a matching, non-matching, or uncertain food matching status. Evidence is rejected when the food does not match, and its automatic propagation is blocked and it enters manual review when the food is uncertain. Evidence is allowed to enter the fact fusion process only when both the brand matching status and the food matching status are matched. The process involves: breaking down evidence matched through dual identity into atomic facts associated with evidence identifiers, food attribute types, standardized attribute values, and assertion types; performing attribute-level consistency and conflict detection on atomic facts from different valid sources that point to the same target SKU; identifying stable atomic facts supported by multiple valid sources and not involved in the conflict as stable atomic facts and retaining them; and reviewing conflicting atomic facts along with their corresponding evidence identifiers to avoid covering other valid sources by selecting a single conflicting source; generating an evidence quality status based at least on source traceability, brand matching status, dish matching status, sufficiency of objective evidence, and conflict status; and implementing one of the following evidence gating routes based on the evidence quality status: candidate or approval, manual review, review after retaining stable atomic facts, or rejection. And generate a traceable mapping between the target food product SKU and standard element identifier based solely on atomic facts that have been gated by the evidence, and associate the evidence identifier set, brand matching status, dish matching status, conflict status and review status in the mapping.

2. The method according to claim 1, characterized in that, The target identity anchors include at least two of the following: standard brand logo, brand name, and SKU name, and further include one or more of the following: brand alias, series name, official description, and source context.

3. The method according to claim 1, characterized in that, Brand identity matching includes: standardizing the characters of brand names in the evidence, mapping confirmed official abbreviations, Chinese names, English names, or brand aliases to a unified standard brand identifier, and removing non-brand identity parts such as city, shopping mall, or store type from store names; generating a brand matching status when the evidence brand name corresponds to the target brand's standard name or confirmed alias, or when the evidence comes from an official context confirmed to belong to the target brand and there are no conflicting brand identifiers; generating a brand mismatch status when the evidence clearly points to another brand; and generating a brand uncertain status in other cases where unique identification is not possible.

4. The method according to claim 1, characterized in that, The dish identity matching process involves standardizing the SKU name and series name respectively, and generating a dish matching status when at least one of the following conditions is met: the standardized complete SKU name is consistent and there is no conflict in the series information; the core product name is consistent and the series name is consistent; or the core product name is consistent, the evidence product description and the existing official description of the target SKU contain the same identity information and there is no conflict with similar SKUs of the same brand.

5. The method according to claim 4, characterized in that, When evidence clearly corresponds to another SKU or series information under the same brand that conflicts with the target SKU, a dish mismatch status is generated; when it cannot be uniquely confirmed that it belongs to the target SKU, a dish uncertainty status is generated.

6. The method according to claim 1, characterized in that, The atomic facts include at least three of the following: evidence identifier, food attribute type, standardized attribute value, assertion type, and source type. Among them, the food attribute type includes at least one of the following: raw material, tea base, milk base, fruit, ingredients, flavor, process, and product form.

7. The method according to claim 1, characterized in that, The conflict detection includes at least the following: when one source explicitly claims to contain the target attribute while another source explicitly claims not to contain the target attribute for the same fact, a conflict is determined; when two sources with confirmed identities give incompatible values ​​for the same attribute that is defined as mutually exclusive, a conflict is determined; if the evidence belongs to different product versions, different SKUs, or the identity cannot be confirmed, the process is transferred to identity or version review instead of directly determining a factual conflict; atomic facts that can be established simultaneously are not determined to be conflicting.

8. The method according to claim 1, characterized in that, The evidence quality status includes at least high quality, insufficient evidence, conflicting evidence, and rejection; among which, a high quality status is generated when the source is traceable, the brand matches, the dish matches, and there is objective original evidence supporting the target element without conflict. An insufficient evidence state is generated when the identity is basically established but the evidence is insufficient to uniquely support the element of fact; an evidence conflict state is generated when the identity is established but there is a factual conflict among multiple valid sources; and a rejection state is generated when the brand or dish does not match, the source is untraceable, or no valid evidence can be formed.

9. The method according to claim 1, characterized in that, The evidence gating includes at least the following: when the evidence is of high quality and without conflict, it enters the candidate or approval state; when the evidence is insufficient, it enters the manual review; when there is a conflict in the source, the stable atomic facts are retained and the conflicting atomic facts are entered into the review; when the identity does not match, the corresponding evidence is rejected.

10. The method according to claim 1, characterized in that, The traceable mapping includes at least six of the following: SKU identifier, standard brand identifier, standard element identifier, brand matching status, dish matching status, evidence quality status, conflict status, manual review status, evidence identifier set, scope of application, rejection reason, and mapping version.

11. A multi-source food product element identification system, characterized in that, The device includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the method according to any one of claims 1 to 10.

12. An electronic device, characterized in that, It includes a processor and a memory, the processor being configured to perform the method according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer program is stored for implementing the method according to any one of claims 1 to 10.

Images