前述のように、関連するクラスタ ID は、そのクラスタ内のすべての例の他の特徴量に置き換えることができます。この置換により特徴の数が少なくなるため、そのデータに基づいてモデルを保存、処理、トレーニングするために必要なリソースも削減されます。非常に大規模なデータセットの場合、この節約は大幅になります。
たとえば、1 本の YouTube 動画には次のような特徴データが含まれる場合があります。
視聴者の位置情報、時間、ユーザー属性
コメントのタイムスタンプ、テキスト、ユーザー ID
動画タグ
YouTube 動画をクラスタリングすると、この特徴セットが単一のクラスタ ID に置き換えられ、データが圧縮されます。
プライバシーの保護
ユーザーをクラスタ化し、ユーザー ID ではなくクラスタ ID にユーザーデータを関連付けることで、プライバシーをある程度保護できます。たとえば、YouTube ユーザーの視聴履歴に基づいてモデルをトレーニングするとします。ユーザー ID をモデルに渡す代わりに、ユーザーをクラスタ化してクラスタ ID のみを渡すことができます。これにより、個々の視聴履歴が個々のユーザーに関連付けられることがなくなります。プライバシーを保護するには、クラスタに十分な数のユーザーを含める必要があります。
[[["わかりやすい","easyToUnderstand","thumb-up"],["問題の解決に役立った","solvedMyProblem","thumb-up"],["その他","otherUp","thumb-up"]],[["必要な情報がない","missingTheInformationINeed","thumb-down"],["複雑すぎる / 手順が多すぎる","tooComplicatedTooManySteps","thumb-down"],["最新ではない","outOfDate","thumb-down"],["翻訳に関する問題","translationIssue","thumb-down"],["サンプル / コードに問題がある","samplesCodeIssue","thumb-down"],["その他","otherDown","thumb-down"]],["最終更新日 2025-02-25 UTC。"],[[["\u003cp\u003eClustering is an unsupervised machine learning technique used to group similar unlabeled data points into clusters based on defined similarity measures.\u003c/p\u003e\n"],["\u003cp\u003eCluster analysis can be applied to various domains like market segmentation, social network analysis, and medical imaging to identify patterns and simplify complex datasets.\u003c/p\u003e\n"],["\u003cp\u003eClustering enables data compression by replacing numerous features with a single cluster ID, reducing storage and processing needs.\u003c/p\u003e\n"],["\u003cp\u003eIt facilitates data imputation by inferring missing feature data from other examples within the same cluster.\u003c/p\u003e\n"],["\u003cp\u003eClustering offers a degree of privacy preservation by associating user data with cluster IDs instead of individual identifiers.\u003c/p\u003e\n"]]],[],null,["Suppose you are working with a dataset that includes patient information from a\nhealthcare system. The dataset is complex and includes both categorical and\nnumeric features. You want to find patterns and similarities in the dataset.\nHow might you approach this task?\n\n[**Clustering**](/machine-learning/glossary#clustering) is an unsupervised\nmachine learning technique designed to group\n[**unlabeled examples**](https://developers.google.com/machine-learning/glossary#unlabeled_example)\nbased on their similarity to each other. (If the examples are labeled, this\nkind of grouping is called\n[**classification**](https://developers.google.com/machine-learning/glossary#classification_model).)\nConsider a hypothetical patient\nstudy designed to evaluate a new treatment protocol. During the study, patients\nreport how many times per week they experience symptoms and the severity of the\nsymptoms. Researchers can use clustering analysis to group patients with similar\ntreatment responses into clusters. Figure 1 demonstrates one possible grouping\nof simulated data into three clusters.\n**Figure 1: Unlabeled examples grouped into three clusters\n(simulated data).**\n\nLooking at the unlabeled data on the left of Figure 1, you could guess that\nthe data forms three clusters, even without a formal definition of similarity\nbetween data points. In real-world applications, however, you need to explicitly\ndefine a **similarity measure** , or the metric used to compare samples, in\nterms of the dataset's features. When examples have only a couple of features,\nvisualizing and measuring similarity is straightforward. But as the number of\nfeatures increases, combining and comparing features becomes less intuitive\nand more complex. Different similarity measures may be more or less appropriate\nfor different clustering scenarios, and this course will address choosing an\nappropriate similarity measure in later sections:\n[Manual similarity measures](/machine-learning/clustering/kmeans/similarity-measure)\nand\n[Similarity measure from embeddings](/machine-learning/clustering/autoencoder/similarity-measure).\n\nAfter clustering, each group is assigned a unique label called a **cluster ID**.\nClustering is powerful because it can simplify large, complex datasets with\nmany features to a single cluster ID.\n\nClustering use cases\n\nClustering is useful in a variety of industries. Some common applications\nfor clustering:\n\n- Market segmentation\n- Social network analysis\n- Search result grouping\n- Medical imaging\n- Image segmentation\n- Anomaly detection\n\nSome specific examples of clustering:\n\n- The [Hertzsprung-Russell diagram](https://wikipedia.org/wiki/Hertzsprung%E2%80%93Russell_diagram) shows clusters of stars when plotted by luminosity and temperature.\n- Gene sequencing that shows previously unknown genetic similarities and dissimilarities between species has led to the revision of taxonomies previously based on appearances.\n- The [Big 5](https://wikipedia.org/wiki/Big_Five_personality_traits) model of personality traits was developed by clustering words that describe personality into 5 groups. The [HEXACO](https://wikipedia.org/wiki/HEXACO_model_of_personality_structure) model uses 6 clusters instead of 5.\n\nImputation\n\nWhen some examples in a cluster have missing feature data, you can infer the\nmissing data from other examples in the cluster. This is called\n[imputation](https://developers.google.com/machine-learning/glossary/#value-imputation).\nFor example, less popular videos can be clustered with more popular videos\nto improve video recommendations.\n\nData compression\n\nAs discussed, the relevant cluster ID can replace other features for all\nexamples in that cluster. This substitution reduces the number of features and\ntherefore also reduces the resources needed to store, process, and train models\non that data. For very large datasets, these savings become significant.\n\nTo give an example, a single YouTube video can have feature data including:\n\n- viewer location, time, and demographics\n- comment timestamps, text, and user IDs\n- video tags\n\nClustering YouTube videos replaces this set of features with a\nsingle cluster ID, thus compressing the data.\n\nPrivacy preservation\n\nYou can preserve privacy somewhat by clustering users and associating user data\nwith cluster IDs instead of user IDs. To give one possible example, say you want\nto train a model on YouTube users' watch history. Instead of passing user IDs\nto the model, you could cluster users and pass only the cluster ID. This\nkeeps individual watch histories from being attached to individual users. Note\nthat the cluster must contain a sufficiently large number of users in order to\npreserve privacy.\n| **Key terms:**\n|\n| - [clustering](/machine-learning/glossary#clustering)\n| - [example](/machine-learning/glossary#example)\n| - [unlabeled example](/machine-learning/glossary#unlabeled_example)\n| - [classification](/machine-learning/glossary#classification_model)"]]
