The algoirthm is registered under MIT License Case No. 25019. 

The code and data is solely for academic and/or internal research purposes. Please see the Internal Research End User Agreement for Academic and other Non-Profit Research Institutions for details. Any requests should be sent to mksc.pe.replication@gmail.com with a copy of signed agreement.

1.  List of code

The code contains 11 scripts in Python or R divided into construction
(scripts 1-5), evaluation (scripts 6-7), and interpretation (scripts
8-11) of the algorithm. Each script is named after the task it performs.
The main results reported in the paper and online appendices can be
reproduced from scripts 6, 7, and 11.

2.  List of data frames in replication.RData

  -----------------------------------------------------------------------
  Name                 Description
  -------------------- --------------------------------------------------
  activity             Activities detected by the I3D algorithm and their
                       association with engagement scores

  e_space              Average pixel-level engagement scores over space

  e_time               Average pixel-level engagement scores over time

  p_space              Average pixel-level product scores over space

  p_time               Average pixel-level product scores over time

  pe_space             Average pixel-level PE-scores over space

  pe_time              Average pixel-level PE-scores over time

  emotion              Scores for different emotions detected by the FER
                       algorithm and engagement scores

  loss                 Training and validation losses for the 3D CNN

  model_construction   Video engagement data for constructing the 3D CNN

  model_evaluation     Sales panel

  object               Objects and number of instances detected by YOLO
                       algorithm

  own                  Influencer and product features for videos in the
                       sales panel and videos that advertise the
                       influencers' own products

  search               Baidu search index
  -----------------------------------------------------------------------

3.  Variable dictionary

"activity"

Each observation is at activity level.

  -----------------------------------------------------------------------------
  Variable       Description
  -------------- --------------------------------------------------------------
  Estimate       The association between the probability of an activity with
                 engagement score at video-segment (15 seconds) level

  Std..Error     Standard error

  t.value        T value

  Pr\...t..      P value

  act            Name of the activity
  -----------------------------------------------------------------------------

"e_space"

A 224 x 224 matrix of engagement scores at each pixel location averaged
over all videos.

"e_time"

Each observation is at video-second level.

  -----------------------------------------------------------------------
  Variable       Description
  -------------- --------------------------------------------------------
  time           Time in second

  value          Average engagement score at that time point
  -----------------------------------------------------------------------

"p_space"

A 224 x 224 matrix of product scores at each pixel location averaged
over all videos.

"p_time"

Each observation is at video-second level.

  -----------------------------------------------------------------------
  Variable        Description
  --------------- -------------------------------------------------------
  time            Time in second

  value           Average product score at that time point
  -----------------------------------------------------------------------

"pe_space"

A 224 x 224 matrix of product engagement scores at each pixel location
averaged over all videos.

"pe_time"

Each observation is at video-second level.

  -------------------------------------------------------------------------
  Variable       Description
  -------------- ----------------------------------------------------------
  time           Time in second

  value          Average product engagement score at that time point
  -------------------------------------------------------------------------

"emotion"

Each observation is at video-frame level.

  -----------------------------------------------------------------------
  Variable          Description
  ----------------- -----------------------------------------------------
  video_id          Video ID

  time              Time in second

  angry             Anger score

  disgust           Disgust score

  fear              Fear score

  happy             Happiness score

  sad               Sadness score

  surprise          Surprise score

  neutral           Neutral score

  enga_score        Average pixel-level engagement score
  -----------------------------------------------------------------------

"loss"

Each observation is at epoch level.

  -----------------------------------------------------------------------
  Variable           Description
  ------------------ ----------------------------------------------------
  epoch              Epoch number

  Training Loss      Training loss (mean absolute percentage error)

  Validation Loss    Validation loss (mean absolute percentage error)
  -----------------------------------------------------------------------

"model_construction"

Each observation is at video level.

  -----------------------------------------------------------------------
  Variable       Description
  -------------- --------------------------------------------------------
  video_id       Video ID 

  like           Number of likes received by the video

  comment        Number of comments received by the video

  share          Number of shares received by the video
  -----------------------------------------------------------------------

"model_evaluation"

Each observation is at product/video-day level.

  ----------------------------------------------------------------------------
  Variable            Description
  ------------------- --------------------------------------------------------
  video_id            Video ID

  updated_time        Date

  updated_time_rel    Number of days since the first day in the sales panel
                      for each product

  video_posted_time   Posting date of the video

  treated             Whether the video ad has been posted for a product (1
                      for posted, 0 for not yet)

  like                Number of likes received by the video

  comment             Number of comments received by the video

  share               Number of shares received by the video

  len                 Length of the video in second

  e_score             Engagement score constructed by the number of shares
                      (normalized to the interval of [0, 1])

  e_score_like        Engagement score constructed by the number of likes

  e_score_comment     Engagement score constructed by the number of comments

  e_score_unsup       Engagement score constructed using an unsupervised
                      approach

  p_score             Product score of the video (normalized to the interval
                      of [0, 1])

  pe_score            PE-score (normalized to the interval of [0, 1])

  pe_score_like       PE-score constructed by the number of likes (normalized
                      to the interval of [0, 1])

  pe_score_comment    PE-score constructed by the number of comments
                      (normalized to the interval of [0, 1])

  pe_score_unsup      PE-score constructed using an unsupervised approach
                      (normalized to the interval of [0, 1])

  influencer_id       Influencer ID

  gender              Gender of the influencer (0 for female, 1 otherwise)

  fans                Number of followers of the influencer

  order_cnt           Number of video ads the influencer has posted

  avg_play            Average play for the influencer

  influencer_price    Price per video ad for the influencer (in RMB)

  expected_cpm        Expected CPM for the influencer (in RMB)

  taobao_id           Product ID

  rev                 Previous 30-day sales revenue (in RMB)

  search              Baidu search index for the product

  avg_search          Average Baidu search index over time

  price               Product price (in RMB)

  discount            Product price discount (in RMB)

  category            Category of the product
  ----------------------------------------------------------------------------

"object"

Each observation is at object level.

  ----------------------------------------------------------------------------
  Variable       Description
  -------------- -------------------------------------------------------------
  object         Object detected by the YOLO algorithm

  count_high     Number of object instances detected in high-engagement pixels

  count_low      Number of object instances detected in low-engagement pixels

  count_diff     Difference between the number of object instances detected in
                 high-engagement pixels and low-engagement pixels
  ----------------------------------------------------------------------------

"own"

Each observation is at video level.

  --------------------------------------------------------------------------
  Variable       Description
  -------------- -----------------------------------------------------------
  video_id       Video ID

  pe_score       PE-score of the video

  own            Whether the video advertises the influencer's own product

  category       Category of the product

  fans           Number of followers for the influencer

  gender         Gender of the influencer (0 for female, 1 otherwise)

  order_cnt      Number of video ads the influencer has posted

  price          Product price (in RMB)

  discount       Product price discount (in RMB)
  --------------------------------------------------------------------------

"search"

Each observation is at product-day level.

  -----------------------------------------------------------------------
  Variable                       Description
  ------------------------------ ----------------------------------------
  taobao_id                      Product ID

  updated_time                   Date

  search                         Baidu search index
  -----------------------------------------------------------------------