Fitting-Humans-Stories-in-List-Columns_eRum

eRum 2018 Talk

Transcript

1. FITTING HUMANS STORIES IN LIST COLUMNS Cases from an Online Recruitment Platform Omayma Said @OmaymaS
2. The Leading Job Site in EGYPT
3. 19th Century Adolphe Quetelet
4. THE AVERAGE MAN (L’homme Moyen) 19th Century Adolphe Quetelet
5. THE AVERAGE MAN Physical Weight, Height (Body Mass Index)
6. THE AVERAGE MAN Social Marriage
7. The AVERAGE MAN Moral Crimes
8. PERFECTION THE AVERAGE MAN = For Quetelet
9. If an individual at any given epoch of society possessed all the qualities of the AVERAGE MAN, he would represent all that is great, good, or beautiful. “ ” Adolphe Quetelet
10. Who Is The “AVERAGE MAN” in Your Society?
11. Are You Just a Deviant from The “AVERAGE MAN” ?
12. Many Disagree !
13. Now...
14. Now... Tremendous Growth of Data
15. Misuse of SUMMARY STATISTICS
16. Misuse of SUMMARY STATISTICS
18. Misuse of SUMMARY STATISTICS
22. The Leading Job Site in EGYPT
24. What Do We Optimize For? Quality Quantity Relevance Matching Jobs & Job Seekers
25. Let’s talk about DATA KPIs METRICS
26. “The average job seeker applies for N jobs per month” Me:
27. “The average number of applications per job this month is GREAT” Me:
28. What AVERAGE Do You Measure?
29. Who is The AVERAGE Job Seeker?
30. Can We Tell Better STORIES About Our Users?
32. Effective Data Analysis Contextual Understanding + We can tell better stories with….
33. Effective Data Analysis Contextual Understanding Culture Socioeconomic Status Market Dynamics +
34. Effective Data Analysis Contextual Understanding Mindset Workflow Framework/Tools +
35. Effective Data Analysis Mindset Workflow Framework/Tools + Contextual Understanding Culture Socioeconomic Status Market Dynamics
36. Effective Data Analysis Effective Data Analysis Contextual Understanding Contextual Understanding Better Stories = +
37. Effective Data Analysis Effective Data Analysis Contextual Understanding Contextual Understanding Actionable Insights = +
38. Framework/Tools https://speakerdeck.com/hadley/tidyverse Compatible Packages +
39. Main Concepts The Tidyverse Let’s focus on
40. Tidy Data Three Main Concepts by: @_inundata & @jcheng
41. A variable in a column An observation in a row Tidy your data And here you go! Tidy Data Three Main Concepts [tibble, tidyr, dplyr, and friends ]
42. Data comes from different SOURCES And more...
43. Data comes in different FORMATS And more...
44. Data comes in different FORMATS DATAFRAME (TIBBLE) Read Tidy
45. user job_id job_title company application_date Sara A1234 Software Developer Company A 2017-01-02 Sara A1568 Senior Software Engineer Company B 2017-03-02 Sara A1590 Software Engineer Company C 2017-03-03 …... ….. …. …. …. Omar A1234 Software Developer Company A 2017-01-03 Omar A1580 Android Developer Company C 2017-01-20 ….. …. …. …. ….. Tidy Data
46. Nested Data Three Main Concepts
47. Nested Data One row per group Instead of One row per observation Three Main Concepts [tidyr ]
48. user job_id job_title company application_date Sara A1234 Software Developer Company A 2017-01-02 Sara A1568 Senior Software Engineer Company B 2017-03-02 Sara A1590 Software Engineer Company C 2017-03-03 …... ….. …. …. …. Omar A1234 Software Developer Company A 2017-01-03 Omar A1580 Android Developer Company C 2017-01-20 ….. …. …. …. ….. user_data %>% group_by(user) %>% nest(.key = “applications”) user applications Sara <Tibble [3 x 4]> Omar <Tibble [2 x 4]> …. …... Nested Data
49. user job_id job_title company application_date Sara A1234 Software Developer Company A 2017-01-02 Sara A1568 Senior Software Engineer Company B 2017-03-02 Sara A1590 Software Engineer Company C 2017-03-03 …... ….. …. …. …. Omar A1234 Software Developer Company A 2017-01-03 Omar A1580 Android Developer Company C 2017-01-20 ….. …. …. …. ….. job_data %>% group_by(job_id) %>% nest(.key = “applications”) job_id applications A1234 <Tibble [2 x 4]> A1568 <Tibble [30 x 4]> A1590 <Tibble [100 x 4]> A1580 <Tibble [120 x 4]> Nested Data
50. Functional Programming Three Main Concepts
51. Functional Programming Handle iteration problems powerfully and emphasize the actions rather than the objects Three Main Concepts [purrr ]
52. Let’s store models in columns job_id applications app_count A5638 <tibble [362 x 27]> 362 A8957 <tibble [110 x 27]> 110 ….. ….. ….. job_app_data<- job_app_data %>% mutate(glm_model = map(app_data, ~ glm(viewed ~ app_day, data = .x, family = binomial)))
53. job_id applications app_count glm_model A5638 <tibble [362 x 27]> 362 <S3: glm> A8957 <tibble [110 x 27]> 110 <S3: glm> ….. ….. ….. …. job_app_data<- job_app_data %>% mutate(glm_model = map(app_data, ~ glm(viewed ~ app_day, data = .x, family = binomial))) Let’s store models in columns
54. user_data <- user_data %>% mutate(common_jobs = map2(applications, preferences, ~intersect(.x[[“job_title”],.y[[“job_title”]]) Iterate and answer more questions user applications preferences Sara <tibble [2 x 10]> <tibble [4 x 10]> Omar <tibble [2 x 15]> <tibble [2 x 10]> ….. ….. ….
55. user applications preferences common_jobs Sara <tibble [2 x 10]> <tibble [4 x 10]> <chr [2]> Omar <tibble [2 x 15]> <tibble [2 x 10]> <chr [0]> ….. ….. …. Iterate and answer more questions user_data <- user_data %>% mutate(common_jobs = map2(applications, preferences, ~intersect(.x[[“job_title”],.y[[“job_title”]])
56. Let’s Look Closer !
57. Problem Shortage in applications for certain Software Development jobs Overall growth and good KPIs
58. Problem Shortage in applications for certain Software Development jobs Dissatisfied Employers
59. Problem Flagged by different sources Shortage in applications for certain Software Development jobs
60. Problem Masked by high-level metrics Shortage in applications for certain Software Development jobs
62. Talent Shortage Hypotheses What if we just have a small pool of job seekers who are interested in the affected jobs?
63. Hypotheses Irrelevant Jobs Maybe employers are not catching up with the global trends or job seekers aspirations!
64. Hypotheses Hidden Jobs What if some jobs do not get enough exposure in the search/recommendation pages?
65. st The Job’s Side Investigation
66. What about applications details per job? The Job’s Side
67. The Job’s Side Job applications details
68. What about iOS job applications? The Job’s Side
69. Job Applications Growth over time iOS Developers Jobs
70. What happens to job posts on day X? Day 7 iOS Developers Jobs
71. What is special about these jobs? Mobile Developer (iOS, Android) iOS Developers Jobs
72. What about the rest? iOS Developers Jobs
73. More with Shiny... *Sample of Wuzzuf Job Posts
74. nd The Job Seeker’s Side Investigation
75. How do job seekers fill their profiles? The Job Seeker’s Side tidytext
76. The Job Seeker’s Side How do job seekers fill their profiles? Details of job seeker’s keywords
77. What about the repetition in the extracted keywords? The Job Seeker’s Side
78. The Job Seeker’s Side What about the repetition in the extracted keywords? Summaries from Job Seeker's Keywords
79. Which jobs match each user’s profile? The Job Seeker’s Side solrium
80. Which jobs match each user’s profile? The Job Seeker’s Side
81. Which jobs match each user’s profile? The Job Seeker’s Side Recommended Jobs Details
82. What ACTIONS Did This Analysis Trigger?
83. Talent Shortage Recommended Actions - Acquire more senior developers - Activate the existing developers - Support the community
84. Irrelevant Jobs - Advise employers about the market Recommended Actions - Revisit preference-based matching
85. Hidden Jobs - Revisit text fields indexing - Tune field weights for scoring - Improve mail recommendation Recommended Actions
86. Main Concepts Tidy Data Nested Data Functional Programming Effective Data Analysis Contextual Understanding + = Actionable Insights @OmaymaS
87. FITTING HUMANS STORIES IN LIST COLUMNS Cases from an Online Recruitment Platform Omayma Said @OmaymaS