{"id":1021573,"date":"2025-10-09T10:22:34","date_gmt":"2025-10-09T09:22:34","guid":{"rendered":"https:\/\/www.artefact.com\/?post_type=blog&p=1021573"},"modified":"2025-12-02T09:17:24","modified_gmt":"2025-12-02T09:17:24","slug":"enriching-the-diy-experience-how-adeo-uses-ai-to-connect-content-and-knowledge","status":"publish","type":"blog","link":"https:\/\/www.artefact.com\/br\/blog\/enriching-the-diy-experience-how-adeo-uses-ai-to-connect-content-and-knowledge\/","title":{"rendered":"Enriquecendo a experi\u00eancia DIY: Como a ADEO usa o AI para conectar conte\u00fado e conhecimento"},"content":{"rendered":"

<\/span><\/a><\/span>

Read the article on<\/p>\n<\/div>

<\/a><\/span><\/div>
<\/div><\/div><\/div><\/div>
<\/article>
<\/div>
<\/div><\/div><\/div>
<\/div><\/p>\n

<\/h2>\n

Context<\/h2>\n

ADEO has developed an extensive\u00a0Knowledge Graph<\/strong> that encompasses its entire product catalog. Simultaneously, the company publishes a wealth of DIY articles on its website. However, these articles remain disconnected from the Knowledge Graph, preventing us from accurately identifying which products or entities within the taxonomy are referenced in the content. By linking these articles to the Knowledge Graph, ADEO could significantly elevate the user experience through smarter search capabilities, personalized recommendations, and more engaging, enriched content.<\/p>\n

This initiative marks the latest chapter in a successful and enduring collaboration between Adeo, Google, and Artefact. Building on a foundation of shared expertise in data, retail, and cutting-edge technology, this project represents a natural evolution in our journey to innovate the digital retail landscape. The strategic alignment with Google has been instrumental in providing the tools and infrastructure necessary to tackle this ambitious endeavor.<\/p>\n

The Cornerstone: Adeo\u2019s Knowledge Graph & DIY Article Potential<\/h2>\n

At the heart of this project lies Adeo\u2019s robust Knowledge Graph \u2014 a sophisticated graph database housing the company\u2019s taxonomy \u2014 which is a structured way of classifying and categorizing information. This network of interconnected data points, currently comprising around\u00a0500,000 relations<\/strong> with\u00a023,000 unique subjects<\/strong>,\u00a041 predicates<\/strong>, and\u00a0225,000 objects<\/strong>, represents a wealth of information about products, categories, and their relationships. Here are simple examples of relations you might find in this knowledge graph:<\/p>\n

\"\"<\/p>\n

Examples of entities and relationships<\/em><\/p>\n

However, a significant portion of valuable information resides within the numerous\u00a0Do-It-Yourself (DIY) articles<\/strong> published on the Leroy Merlin website. These articles, rich with practical advice and instructions, often mention entities already present within Adeo\u2019s Knowledge Graph. The challenge? There was\u00a0no automated way<\/strong> to identify these mentions and forge the crucial links between the textual content and the structured knowledge.<\/p>\n

Bridging this gap unlocks significant\u00a0business value<\/strong>, especially within the context of an ongoing AI and Gen AI transformation. By automatically extracting entities from articles and other textual data and linking them in the Knowledge Graph, and thus by enriching it, we can:<\/p>\n

    \n
  • Improve Search Relevance:<\/strong>\u00a0Enable\u00a0semantic search<\/strong>, allowing users to find articles based on the underlying concepts rather than just keywords.<\/li>\n
  • Enhance Product Recommendations:<\/strong>\u00a0Understand the entities discussed in an article to recommend relevant products, tools, and materials directly to the reader.<\/li>\n
  • Enrich and Personalize Content:<\/strong>\u00a0Dynamically enrich articles with links to relevant entities in the Knowledge Graph, providing users with deeper context and related information.<\/li>\n<\/ul>\n

    Navigating the Landscape: NER & NEL with LLMs<\/h2>\n

    The task at hand \u2014 identifying and linking mentions of entities within text to a predefined knowledge base \u2014 falls under the well-established domains of\u00a0Named Entity Recognition (NER)<\/strong>\u00a0and Named Entity Linking (NEL)<\/strong>. Traditionally, high performance required training specialized models on large, labeled datasets. While powerful NER\/NEL models exist, their data-intensive nature\u00a0presented<\/strong>\u00a0a challenge for our rapid deployment needs.<\/p>\n

    Therefore, we chose a\u00a0different approach<\/strong>: leveraging the power of\u00a0Large Language Models (LLMs)<\/strong>\u00a0to build our extraction pipeline. While LLMs require little to no task-specific training data \u2014 allowing for faster implementation and iteration \u2014 they still demand\u00a0annotated data<\/strong>\u00a0for evaluation. To this end, the Adeo team built a comprehensive\u00a0validation set<\/strong>, which required significant human effort and deep business expertise. This dataset is essential for reliably measuring the pipeline\u2019s performance.<\/p>\n

    Our primary goal wasn\u2019t perfect accuracy out of the gate. Instead, we focused on creating a functional pipeline to provide\u00a0pre-annotated text<\/strong> to human labelers. This significantly accelerates the annotation process, making future fine-tuning of specialized models much more efficient.<\/p>\n

    Our Innovative Two-Stage Model<\/h2>\n

    To tackle NER and NEL, we developed a robust two-stage pipeline<\/p>\n

    \"\"<\/p>\n

    The two-tiered NER\/NEL pipeline<\/em><\/p>\n

    1. Named Entity Recognition (NER): Spotting candidate entities<\/h2>\n

    This stage identifies mentions of relevant entities within DIY articles using an LLM. We handle article length with Text Chunking:<\/strong>\u00a0long articles are split into manageable chunks (500 words) for consistent LLM context and better performance. Our NER process uses a dual-level strategy:<\/p>\n

      \n
    • Local Entities:<\/strong> For context-specific mentions, each 500-word chunk undergoes double pass extraction for refinement ( kind of Chain of Thoughts<\/em>\u00a0) using an LLM. Results from all chunks are then combined.<\/li>\n
    • Global Entities:<\/strong> For overarching themes, the full text is processed (again with double extraction using an LLM) for comprehensive coverage.<\/li>\n<\/ul>\n

      This two-tiered approach ensures we capture both granular details and broad concepts effectively.<\/p>\n

      2. Named Entity Linking (NEL): Connecting the Dots to the Knowledge Graph<\/h2>\n

      Once entities are extracted, NEL disambiguates and links them to the most relevant Knowledge Graph entry. This involves:<\/p>\n

      \ud83e\udd1d\u00a0Candidate Generation<\/strong><\/p>\n

      For each extracted entity, we generate potential matches from the KG using a vector store and text embeddings. Only the most semantically similar candidates are kept. We used GCP\u00a0text-multilingual-embedding-002<\/em>\u00a0model with a vector database for this task.<\/p>\n

      To illustrate this, imagine the NER stage extracts the candidate entity \u201clightweight canvas gloves\u201d from a text snippet:<\/em><\/p>\n

      \u201c[\u2026] you can choose\u00a0<\/em>lightweight canvas gloves.\u00a0If you work with your hands in the soil [\u2026]\u201d.<\/em><\/p>\n

      In the Candidate Generation step, the system retrieves potential matches from the Knowledge Graph based on semantic similarity. This might yield a ranked list of candidates such as \u201cdisposable gloves\u201d (Rank 1), \u201cwork gloves\u201d (Rank 2), \u2026, \u201cgardening gloves\u201d (Rank 9), and \u201cglass handling gloves\u201d (Rank 10), among others.<\/em><\/p>\n

      \ud83e\udde0\u00a0Semantic Reranking<\/strong><\/p>\n

      Shortlisted candidates are reranked by an LLM analyzing the entity\u2019s context in the article. Only the top match proceeds. We found 25 candidates to be the optimal number for reranking.<\/p>\n

      \n
      \n

      Continuing our example, the LLM would now analyze the surrounding text \u201c\u2026If you work with your hands in the soil\u2026\u201d and use this context to rerank the candidates. Due to the mention of working with soil, \u201cgardening gloves\u201d would likely be promoted to the top of the list as the most semantically relevant candidate.<\/em><\/p>\n

      \ud83c\udf33\u00a0Hierarchical Ranking<\/strong><\/p>\n

      The selected candidate is positioned within the KG\u2019s hierarchy. Another LLM can either keep the selection or replace it with a more suitable parent, child, or sibling based on context. A hierarchical reranking threshold of 100 ensures the full hierarchy is considered.<\/p>\n

      Consider the following simplified hierarchy in the Knowledge Graph:<\/p>\n

      \"\"<\/p>\n

      In this step, the system verifies if \u201cgardening gloves\u201d is the most appropriate level of specificity. While it\u2019s a good match in our example, if the context had been broader, simply mentioning the need for hand protection without the gardening context, the hierarchical ranking might promote the ancestor entity \u201cgloves\u201d and link it to the corresponding KG entry.<\/em><\/p>\n

      \"\"<\/p>\n

      This multi-step NEL process ensures accurate and meaningful anchoring within the Knowledge Graph.<\/p>\n

      Measuring Success: Our Evaluation Methodology<\/h2>\n

      To ensure the effectiveness of our Knowledge Graph enrichment pipeline for Leroy Merlin\u2019s DIY articles, we implemented a robust evaluation against a carefully built\u00a0ground truth dataset <\/strong>containing entities from the Adeo knowledge graph.<\/p>\n

      This evaluation specifically focuses on the pipeline\u2019s ability to identify and link four key entity classes: ProductSet, HomeSpace, DIYActivity, and Color, at both global and local levels within the articles:<\/p>\n

        \n
      1. ProductSet:<\/strong>\u00a0These are tools, materials, or purchasable products used for home improvement, gardening, or DIY tasks.\u00a0Examples<\/strong>: Concrete grinder, Air-to-air heat pump, Gardening apron, Desk lamp, Smart thermostat<\/li>\n
      2. HomeSpace:<\/strong>\u00a0These represent areas or rooms in a home or garden where DIY activities typically occur.\u00a0Examples<\/strong>: Garage, Garden, Kitchen, Bathroom, Balcony<\/li>\n
      3. DIYActivity:\u00a0<\/strong>These are the tasks or operations related to Do-It-Yourself and home improvement.\u00a0Examples<\/strong>: Painting, Installation, Cleaning, Gardening, Insulation work<\/li>\n
      4. Color:<\/strong>\u00a0This category includes any mentioned color or shade.\u00a0Examples<\/strong>: Creamy white, Teal blue, Light grey, Matt black, Bright yellow<\/li>\n<\/ol>\n

        Evaluating the Full Pipeline (NER & NEL)<\/h2>\n

        We assessed overall performance using:<\/p>\n

          \n
        • Precision:<\/strong>\u00a0Correctly identified & linked entities \/ all identified & linked.<\/li>\n
        • Recall:<\/strong>\u00a0Correctly identified & linked entities \/ all actual entities.<\/li>\n
        • F1 Score:<\/strong>\u00a0A balanced measure of precision and recall.<\/li>\n
        • Fuzzy Match Metrics (distances 1, 2, 3):<\/strong>\u00a0We score errors by their hierarchical distance from the true label: distance 1 for direct neighbors, distance 2 for the next level, etc. A wrong prediction still \u201cpasses\u201d if it lies within the allowed radius, capturing near-misses more fairly.<\/li>\n<\/ul>\n

          \"\"<\/p>\n

          Evaluation using a Fuzzy Metric<\/em><\/p>\n<\/div>\n

          Evaluating NER:<\/strong>\u00a0We compared stemmed extracted entities with stemmed ground truth (case-insensitive). Our NER intentionally over-extracts for high recall.<\/p>\n

          Evaluating NEL:\u00a0<\/strong>Assuming perfect NER, we focused on the accuracy of the linking process using the same metrics as the full pipeline, including fuzzy matching.<\/p>\n

          Key Findings: Promising Results & Growth Areas<\/h2>\n

          Here are the performance metrics of our pipeline<\/p>\n

          Full Pipeline (Exact Match)<\/h3>\n

          \"\"<\/p>\n

          Performance metrics of the NER\/NEL pipeline (Exact Match)<\/em><\/p>\n<\/div>\n

            \n
          • Global Entities:<\/strong>\u00a0Strong precision, lower recall (balanced F1).<\/li>\n
          • Local Entities:<\/strong>\u00a0Varied performance.\u00a0ProductSet<\/strong>\u00a0(key category) showed a solid balance (Precision: 58.9%, Recall: 61.74%, F1: 60.29%).\u00a0Color<\/strong>\u00a0also performed well.\u00a0HomeSpace<\/strong>\u00a0needs improvement in precision.<\/li>\n<\/ul>\n

            Full Pipeline (Fuzzy Match)<\/h3>\n

            \"\"<\/p>\n

            Performance using different fuzzy metrics<\/em><\/p>\n

            Fuzzy metrics improve significantly as the distance increases. This clearly shows that predictions considered incorrect in exact match are still relatively close to the actual value within the graph hierarchy.<\/p>\n

            NER:<\/h3>\n

            As expected, we achieved\u00a0high recall<\/strong>\u00a0but lower precision due to our over-extraction strategy.<\/p>\n

            NEL:<\/h3>\n

            The NEL component effectively\u00a0refined entity linking<\/strong>\u00a0\ud83d\udd17 after NER.<\/p>\n

            Conclusion: Building a Smarter DIY Ecosystem<\/h2>\n

            This project marks a significant step in using AI to enrich the DIY experience on Leroy Merlin\u2019s website. By successfully building a pipeline to link DIY articles to Adeo\u2019s Knowledge Graph, we\u2019ve laid the groundwork for smarter search, personalized recommendations, and richer content.<\/p>\n

            While initial results are promising (especially for ProductSet), we\u2019ve identified areas for optimization, like improving HomeSpace precision. Our decision to use LLMs for rapid initial annotation has been a valuable strategy, accelerating data generation for future model training and improvements.<\/p>\n

            The ongoing collaboration between Adeo, Google, and Artefact continues to drive retail innovation. This Knowledge Graph enrichment initiative showcases the power of combining domain expertise with cutting-edge AI to create a more intuitive and valuable experience for DIY enthusiasts. As our pipeline evolves with further refinements and potentially more advanced models like Gemini 2.5 Pro, the connection between content and knowledge will only strengthen, further empowering Leroy Merlin\u2019s customers in their home improvement journeys.<\/p>\n","protected":false},"excerpt":{"rendered":"

            A otimiza\u00e7\u00e3o do sortimento \u00e9 um processo cr\u00edtico no varejo que envolve a sele\u00e7\u00e3o do mix ideal de produtos para atender \u00e0 demanda do consumidor, levando em conta as diversas restri\u00e7\u00f5es log\u00edsticas envolvidas. Os varejistas precisam ter certeza de que est\u00e3o oferecendo os produtos certos, nas quantidades certas e no momento certo. Ao aproveitar o data e as percep\u00e7\u00f5es do consumidor, os varejistas podem tomar decis\u00f5es informadas sobre quais itens estocar, como gerenciar o estoque e quais produtos priorizar com base nas prefer\u00eancias do cliente, nas tend\u00eancias sazonais e nos padr\u00f5es de vendas.<\/p>","protected":false},"featured_media":1021576,"parent":0,"template":"","meta":{"_acf_changed":false,"ep_exclude_from_search":false},"blog-category":[21939,2995,22052],"blog-language":[2991,2993],"class_list":["post-1021573","blog","type-blog","status-publish","has-post-thumbnail","hentry","blog-category-medium","blog-category-ai-technology","blog-category-retail","blog-language-en","blog-language-fr"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog\/1021573","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog"}],"about":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/types\/blog"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/media\/1021576"}],"wp:attachment":[{"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/media?parent=1021573"}],"wp:term":[{"taxonomy":"blog-category","embeddable":true,"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog-category?post=1021573"},{"taxonomy":"blog-language","embeddable":true,"href":"https:\/\/www.artefact.com\/br\/wp-json\/wp\/v2\/blog-language?post=1021573"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}