{
  "id": 324232,
  "title": "Some comments on data quality",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/324232",
  "author_name": "KacperŁukawski",
  "post_date": "2022-05-10T16:23:17.879000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Big congratulations to all the winners! Just out of curiosity, did anybody focus a bit more on the data quality? How did you approach the errors? I probably put too much effort into that, as I realized there are some articles with definitely wrong attribute values.</p>\n<p>Just to give some examples, the following articles have been assigned a visually wrong label <code>product_type_name = \"Bra\"</code>:</p>\n<table>\n<thead>\n<tr>\n<th>article_id</th>\n<th>photo</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>0526980001</td>\n<td><img src=\"https://i.imgur.com/Km5V5Ga.jpeg\" alt=\"\"></td>\n</tr>\n<tr>\n<td>0517921001</td>\n<td><img src=\"https://i.imgur.com/ideOMHB.jpg\" alt=\"\"></td>\n</tr>\n<tr>\n<td>0606698001</td>\n<td><img src=\"https://i.imgur.com/6prtlxe.jpg\" alt=\"\"></td>\n</tr>\n</tbody>\n</table>\n<p>I even created a simple service using ResNet18 embeddings to search for visually similar images with additional filtering based on attributes derived from <code>articles.csv</code>. If anybody wants to try: <a href=\"https://h-and-m.qdrant.tech/\" target=\"_blank\">https://h-and-m.qdrant.tech/</a> (<em>disclaimer: this is done using <a href=\"https://qdrant.tech/\" target=\"_blank\">Qdrant</a> and I work there for a few days</em>). </p>",
  "messages": [
    {
      "id": 1783738,
      "postDate": "2022-05-10T16:23:17.880Z",
      "content": "<p>Big congratulations to all the winners! Just out of curiosity, did anybody focus a bit more on the data quality? How did you approach the errors? I probably put too much effort into that, as I realized there are some articles with definitely wrong attribute values.</p>\n<p>Just to give some examples, the following articles have been assigned a visually wrong label <code>product_type_name = \"Bra\"</code>:</p>\n<table>\n<thead>\n<tr>\n<th>article_id</th>\n<th>photo</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>0526980001</td>\n<td><img src=\"https://i.imgur.com/Km5V5Ga.jpeg\" alt=\"\"></td>\n</tr>\n<tr>\n<td>0517921001</td>\n<td><img src=\"https://i.imgur.com/ideOMHB.jpg\" alt=\"\"></td>\n</tr>\n<tr>\n<td>0606698001</td>\n<td><img src=\"https://i.imgur.com/6prtlxe.jpg\" alt=\"\"></td>\n</tr>\n</tbody>\n</table>\n<p>I even created a simple service using ResNet18 embeddings to search for visually similar images with additional filtering based on attributes derived from <code>articles.csv</code>. If anybody wants to try: <a href=\"https://h-and-m.qdrant.tech/\" target=\"_blank\">https://h-and-m.qdrant.tech/</a> (<em>disclaimer: this is done using <a href=\"https://qdrant.tech/\" target=\"_blank\">Qdrant</a> and I work there for a few days</em>). </p>",
      "rawMarkdown": "Big congratulations to all the winners! Just out of curiosity, did anybody focus a bit more on the data quality? How did you approach the errors? I probably put too much effort into that, as I realized there are some articles with definitely wrong attribute values.\n\nJust to give some examples, the following articles have been assigned a visually wrong label `product_type_name = \"Bra\"`:\n| article_id | photo |\n| --- | --- |\n| 0526980001 | ![](https://i.imgur.com/Km5V5Ga.jpeg) |\n| 0517921001 | ![](https://i.imgur.com/ideOMHB.jpg) |\n| 0606698001 | ![](https://i.imgur.com/6prtlxe.jpg) |\n\nI even created a simple service using ResNet18 embeddings to search for visually similar images with additional filtering based on attributes derived from `articles.csv`. If anybody wants to try: https://h-and-m.qdrant.tech/ (*disclaimer: this is done using [Qdrant](https://qdrant.tech/) and I work there for a few days*). \n",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1783738": "Big congratulations to all the winners! Just out of curiosity, did anybody focus a bit more on the data quality? How did you approach the errors? I probably put too much effort into that, as I realized there are some articles with definitely wrong attribute values.\n\nJust to give some examples, the following articles have been assigned a visually wrong label `product_type_name = \"Bra\"`:\n| article_id | photo |\n| --- | --- |\n| 0526980001 | ![](https://i.imgur.com/Km5V5Ga.jpeg) |\n| 0517921001 | ![](https://i.imgur.com/ideOMHB.jpg) |\n| 0606698001 | ![](https://i.imgur.com/6prtlxe.jpg) |\n\nI even created a simple service using ResNet18 embeddings to search for visually similar images with additional filtering based on attributes derived from `articles.csv`. If anybody wants to try: https://h-and-m.qdrant.tech/ (*disclaimer: this is done using [Qdrant](https://qdrant.tech/) and I work there for a few days*). \n"
  }
}