Changes between Version 5 and Version 6 of ClassificationOfUsersForAds


Ignore:
Timestamp:
Jul 22, 2017, 1:02:56 AM (2 years ago)
Author:
psantos
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • ClassificationOfUsersForAds

    v5 v6  
    33[[PageOutline(2-4)]]
    44
    5 == Project Details ==
    6 
    7 The proposed system can be a general user classification
    8 mechanism that would allow us to scale app sponsorship for
    9 groups of users with similar interests. This system could be applied to
    10 the organic recommendations by the Editorial and Community teams, too.
    11 The system is partially automated, since it allows us to understand each
    12 user group qualitatively and manually manage the list of apps for
    13 each group over time. Part of these recommendations can be
    14 automated based on simple rules such as app categories, trending apps
    15 within each group and most downloaded but not yet installed apps per
    16 group. It is considered that this combined marketing research/data
    17 science approach is the best solution to implement a recommendation
    18 system in a very short term. The solution is also scalable and can be
    19 tweaked and improved over time.
    20 The proposed system can be constructed based on a simple three stage
    21 data mining process.
    22 
    235=== Purpose of this project ===
    24 Create a general user classification mechanism that would allow us to scale app recommendations for
    25 groups of users with similar interests. This system could be applied to
    26 the organic recommendations by the Editorial and Community teams,
    27 but also to the Advertising/Monetisation processes.
     6Create a general user classification mechanism that would allow us to scale app recommendations for groups of users with similar interests. This system could be applied to the organic recommendations by the Editorial and Community teams, but also to the Advertising/Monetisation processes.
    287
    298
    30 === Project Description ===
     9
     10== Project Details ==
     11Develop a user classification
     12mechanism that will allow us to scale app sponsorship and app recommendations for
     13groups of users with similar interests. This process will have consist in four phases: [[BR]]
     14{{{1.Identify Most Important Targets
     15{{{This will be achieved by using a combined marketing research/data mining approach. First, gather a sample of users, with all the apps installed per user. Multidimensional Scaling + Clustering can be applied on the sample. The clustering results need to be interpreted using brainstorming sessions with multidisciplinary teams, to construct realistic and rich profiles.}}} }}}[[BR]]
     16{{{2.Build Predictive Algorithm
     17{{{A very simple and computationally cheap algorithm (K-Nearest Neighbors) can be applied on the entire Hadoop Data lake.}}} }}}
     18
     19{{{3.Implement UI Interface Filter per Target}}}
     20{{{4.Plan Content Targeting for Each Group Based on General App Categories.}}}
     21
    3122
    3223