%%{init: {'theme': 'base', 'themeVariables': { 'fontFamily': 'monospace', "fontSize":"16px", "darkMode":false}}}%%
flowchart TB
raw[("<code>data/dating_services.csv</code>")] -->|"i. Extract"| users["User count<br/>from <code>registered_users</code>"]
users -->|"ii. Extract"| year["Year from<br/>the leftover text"]
year -->|"iii. Categorize"| free["Yes/No/Maybe from<br/><code>free</code> columns"]
free -->|"iv. Filter"| filt["Drop missing counts<br/>and duplicate names"]
filt -->|"v. Save"| clean[("<code>data/dating_services_clean.csv</code>")]
style raw fill:#FFFFFF,color:#000000,stroke:#333,stroke-width:1px
style clean fill:#FFFFFF,color:#000000,stroke:#333,stroke-width:1px
style users fill:#4CBB9D,color:#FFFFFF,rx:5,ry:5
style year fill:#4CBB9D,color:#FFFFFF,rx:5,ry:5
style free fill:#4CBB9D,color:#FFFFFF,rx:5,ry:5
style filt fill:#4CBB9D,color:#FFFFFF,rx:5,ry:5
Clean: dating services
In the CSV from scrape_dating.qmd, the registered-user count and the reporting year are both embedded in one string, like "30,000,000 as of 2006[2]". The user count always has thousands separators, so matching a comma-grouped number pulls it out first; whatever’s left over is where the year lives. The “Free” and “All features free” columns are full sentences; we keep the first word (Yes/No/Maybe) as a clean category and leave the original text in place.
library(dplyr)
library(readr)
library(stringr)
dating_clean <- read_csv("data/dating_services.csv") |>
mutate(
users_raw = str_extract(registered_users, "\\d{1,3}(?:,\\d{3})+"),
registered_users_n = as.numeric(str_remove_all(users_raw, ",")),
users_year = as.integer(str_extract(str_remove(registered_users, fixed(users_raw)), "\\d{4}")),
free_clean = str_extract(free, regex("^(Yes|No|Maybe)", ignore_case = TRUE)),
all_features_free_clean = str_extract(all_features_free, regex("^(Yes|No|Maybe)", ignore_case = TRUE))
) |>
select(-users_raw) |>
filter(!is.na(registered_users_n)) |>
distinct(name, .keep_all = TRUE)write_csv(dating_clean, "data/dating_services_clean.csv")
dating_clean
## # A tibble: 26 × 14
## name description owner registered_users free all_features_free hetero poly
## <chr> <chr> <chr> <chr> <chr> <chr> <chr> <chr>
## 1 Adul… Members ca… <NA> 30,000,000 as o… No "No, free users … Yes Yes
## 2 Anas… Connects W… <NA> 4,000,000 as of… No "No" Yes ?
## 3 Ashl… Members ar… <NA> 60,000,000 as o… Yes "No; Free users … Yes Yes
## 4 Badoo Available … Bumb… 500,000,000 as … Yes "No; Charge for … Yes ?
## 5 Beau… Dating web… <NA> 800,000 as of 2… No "No; free profil… Yes ?
## 6 Bumb… Dating app… Bumb… 22,000,000 as o… Yes "No" Yes ?
## 7 Chyr… Dating app… Gree… 1,000,000 downl… Yes "No; Free users … Yes No
## 8 Chri… Dating sit… Spar… 12,000,000 as o… No "No" Yes ?
## 9 Date… Dedicated … <NA> 179,000 as of 2… No; … "Yes" ? Part…
## 10 Duol… Open-sourc… Inde… as of 2026 15,0… Yes "No; optional \"… Yes Unkn…
## # ℹ 16 more rows
## # ℹ 6 more variables: free_messaging <chr>, std_notification <chr>,
## # registered_users_n <dbl>, users_year <int>, free_clean <chr>,
## # all_features_free_clean <chr>import pandas as pd
dating = pd.read_csv("data/dating_services.csv")
users_raw = dating["registered_users"].str.extract(r"(\d{1,3}(?:,\d{3})+)")[0]
dating["registered_users_n"] = pd.to_numeric(users_raw.str.replace(",", ""))
leftover = pd.Series(
[r.replace(u, "", 1) if pd.notna(u) else None for r, u in zip(dating["registered_users"], users_raw)],
index=dating.index, dtype="object"
)
dating["users_year"] = pd.to_numeric(leftover.str.extract(r"(\d{4})")[0]).astype("Int64")
dating["free_clean"] = dating["free"].str.extract(r"^(Yes|No|Maybe)", flags=2)[0]
dating["all_features_free_clean"] = dating["all_features_free"].str.extract(r"^(Yes|No|Maybe)", flags=2)[0]
dating_clean = (
dating.dropna(subset=["registered_users_n"])
.drop_duplicates(subset="name")
)
dating_clean.to_csv("data/dating_services_clean.csv", index=False)
dating_clean.head() name ... all_features_free_clean
1 Adult FriendFinder ... No
2 AnastasiaDate ... No
3 Ashley Madison ... No
4 Badoo ... No
5 BeautifulPeople ... No
[5 rows x 14 columns]
stringr has a slight edge
Finding the year means removing each row’s own user count from its text first. str_remove() is vectorized over the pattern, so one call handles every row. The .str methods in pandas take a single pattern for the whole column, so the Python version loops over the rows with zip().
See viz_dating.qmd for the chart built on this.