用戶細分

介紹 (Introduction)

The goal of this analysis was to identify different user groups based on the deals they have availed, using a discount app, in order to re-target them with offers similar to ones they have availed in the past.

該分析的目的是使用折扣應用程序基于他們所獲得的交易來識別不同的用戶組，以便以與他們過去所獲得的類似的報價來重新定位他們。

Machine learning algorithm K-means was used to identify user segments based on their purchase behavior. Here is a 3-D illustration of what algorithm extracted.

機器學習算法K-means用于根據用戶細分的購買行為識別用戶細分。 這是所提取算法的3D圖。

3D image of clusters produced by K-Means, by MuffaddalMuffaddal的K-Means產生的星團的3D圖像

術語： (Terminologies:)

Before going deeper into the analysis, let’s define some keywords being used.

在深入分析之前，讓我們定義一些正在使用的關鍵字。

Deal Avail: When user avails discount using app.Spent: Discounted price user pays while buying an item.Saved: Amount user saved through the app.Brands: Vendors for which discounts are being offered such as Pizza Hut, GreenODeals: Discounts offered to users on different outlets and brands.

交易無效：當用戶使用應用程序享受折扣時。 已用：用戶在購買商品時支付的折扣價。 已保存：通過應用保存的用戶數量。 品牌：為其提供折扣的供應商，例如必勝客，GreenO 交易：為不同商店和品牌的用戶提供折扣。

分析 (Analysis)

資料集 (Data sets)

The behavior data set was extracted from Mixpanel using JQL. Following was used for this analysis

使用JQL從Mixpanel提取行為數據集。 以下用于此分析

Mixpanel Data Set, by MuffaddalMixpanel數據集，作者：Muffaddal

userId: unique id of usersaveAmount: amount saved by user on deal availspentAmount: amount spent by user on deal availbrandName: brand for which deal was availedcount: number of deals availed by user

userId：用戶的唯一ID saveAmount ：用戶在交易有效時所節省的金額costAmount ：用戶在交易有效時所消耗的金額brandName ：已進行交易的品牌數 ：用戶所進行的交易數量

Using the above data set averageSpentAmount, averageSavedAmount and dealAvailCount was calculated for each user as seen below

使用上面的數據集，為每個用戶計算了averageSpentAmount ， averageSavedAmount和dealAvailCount ，如下所示

Average Deal Availed Data set, by MuffaddalMuffaddal的平均交易可用數據集

Machine Learning — K-means ClusteringThe first step of the k-mean algorithm was to find an optimal number of clusters for segmentation. There are a number of methods out there for this purpose, one of which is the elbow method using within-cluster sum square (wcss).

機器學習-K均值聚類 k均值算法的第一步是找到用于分割的最佳聚類數。 為此，存在許多方法，其中一種是使用簇內和平方(wcss)的肘方法。

WCSS for up-to 10 clusters, by MuffaddalMuffaddal的WCSS最多可支持10個集群

Based on the elbow method, 4, 5, and 6 clusters were used to explore the segments and 4 clusters were picked as best for the given data set.

基于彎頭方法，使用4、5和6個聚類來探索這些段，并且對于給定的數據集，最好選擇4個聚類。

R code for K-Means clustering用于K均值聚類的R代碼

I would recommend these courses on Data camp and Coursera if you want to learn more about user clustering and user segmentation.

如果您想了解有關用戶集群和用戶細分的更多信息， 我將在 數據營 和 Coursera 上推薦這些課程 。

K均值提取了哪些細分？ (What Segments K-means extracted?)

Following were average stats of four identified segments:

以下是四個確定的細分市場的平均統計信息：

Average stats of each segment每個細分的平均統計信息 Segments Characteristics細分特征 Graphical Representation of Segments Characteristics, by Muffaddal段特征的圖形表示，按Muffaddal

Users in segment 1 and 2 were high paying users with segment 1 users also had saved equally high per deal(probably availed buy 1 get 1 offers). However, the number of deals availed by these users were less than 2 (i.e. 1.3 and 1.4 respectively).

第1段和第2段的用戶是高薪用戶，第1段的用戶每筆交易也節省了同樣高的費用(可能使用“買一送一”的優惠)。 但是，這些用戶獲得的交易數量少于2(即分別為1.3和1.4)。

On the other hand, segment 3 and segment 4 users spent less and hence, saved less as well. However, segment 4 users had the greatest deal availed per user ratio (on average more than 9 deals availed by each user) in all 4 segments. It was the most converted cohort of users.

另一方面，第3段和第4段的用戶花費較少，因此節省的也較少。 但是，在所有4個細分受眾群中，細分受眾群4用戶的每位用戶交易比例最高(平均每位用戶超過9筆交易)。 這是轉化率最高的用戶群體。

每個細分市場中的用戶總數和交易數量是多少？ (What were the total number of users and the number of deals availed in each segment?)

Here is the total number of users and deals each segment users had availed.

這是每個細分用戶可用的用戶總數和交易數。

Number of users in segments, by Muffaddal細分中的用戶數(按Muffaddal) Number of deals availed, by Muffaddal通過Muffaddal獲得的交易數量

57% of users belonged to segment 3 and only 3% of users were from the most converted segment (i.e segment 4).

57％的用戶屬于第3部分，而只有3％的用戶來自轉化率最高的部分(即第4部分)。

總體用戶支出是多少？ (What were overall users spending?)

Here is the spread of spending by each segment

這是每個細分受眾群的支出分布

Spending of users in each segment, by Muffaddal每個細分領域的用戶支出，按Muffaddal劃分

Some of the users from segment 4 had high spending (yellow dots in segment 4) similar to segment 1 and 2 but segment 3 (which comprise of 57% of the users) didn't go for high spending deals and/or brands at all.

第4部分的一些用戶具有較高的支出(第4部分中的黃點)，類似于第1和第2部分，但第3部分(占57％的用戶)根本沒有進行高支出的交易和/或品牌推廣。

每個細分市場用戶偏好的品牌類型？ (Type of brand each segment users preferred?)

Let’s look at what type of brand these segment users avail to understand any distinction in them.

讓我們看看這些細分用戶可以使用哪種類型的品牌來理解他們之間的任何區別。

Brands users availed, by MuffaddalMuffaddal推薦的品牌用戶

Segment 1 users had availed mix of burger, pizza and fun time, Segment 2 users had availed pizza and segment 3 users had preferred burgers. While Segment 4 users (most converted users) preferred juices and other types of brands.

第1部分用戶使用了漢堡，比薩餅和娛樂時間，第2部分用戶使用了比薩餅，第3部分用戶則選擇了漢堡。 而第4類用戶(轉化最多的用戶)則更喜歡果汁和其他類型的品牌。

每個細分市場都有哪些品牌？ (What brands each segment availed?.)

Here are the top 10 brands these segmented users had availed.

以下是這些細分用戶所使用的十大品牌。

Top 10 Brands Availed by Each Segments, by Muffaddal各細分市場排名前10位的品牌，按Muffaddal列出

Looking at the brands we can comprehend what type of brand and deals these segment users would prefer. Segment 1 & 2 users (high paying users) had availed premium brands such as Sajjad, kababi, Charcoal, California, etc while segment 3 and 4 (low paying users) had mostly opted in for medium to low tier brands.

通過查看品牌，我們可以了解哪些類型的品牌以及這些細分用戶希望的交易。 第1段和第2段用戶(高收入用戶)曾使用過Sajjad，kababi，木炭，加利福尼亞等高級品牌，而第3段和第4段(低收入用戶)則大多選擇了中低檔品牌。

如何運用這些結果？ (How these results can be employed?)

Based on different user segments we can:

根據不同的用戶群，我們可以：

1- Targeted Ads Personalize ads for each segment would increase the conversion rate as users are more likely to convert on specific brands and offers. So, for example, show Sajjad’s ads to users with higher-paying power then to users with low paying power.

1-定位廣告 每個細分受眾群的個性化廣告將提高轉化率，因為用戶更有可能轉化為特定品牌和優惠。 因此，例如，向高支付能力的用戶展示Sajjad的廣告，然后向低支付能力的用戶展示。

2- In-app RecommendationsOptimize the app to recommend deals and discounts within the app that each segment users would be more interested in.

2-應用內推薦優化應用，以推薦每個細分市場用戶更感興趣的應用內交易和折扣。

摘要 (Summary)

To sum up, with data and proper efforts we were able to identify interesting information about users and their liking and were able to strategies how to engage users more based on their preferences.

綜上所述，通過數據和適當的努力，我們能夠識別出有關用戶及其喜好的有趣信息，并能夠根據用戶的喜好來制定如何與用戶進行更多互動的策略。

翻譯自: https://towardsdatascience.com/user-segmentation-based-on-purchase-history-490c57402d53

用戶細分

總結

以上是生活随笔為你收集整理的用户细分_基于购买历史的用户细分的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。