Icon

Project

Discovering Point of Interest with K-Means

Main words in the center (Reporting)

Photo identity and duplicates

Result: 3,990 photos appear once, 5 appear twice, and 200 photos are copied 254 times.

Check & Remove Duplication

row ID is a line number, different on every copy of the same photo. It must be removed before duplicate detection. The photo identifier we keep is id_photo.

Missing Values

GPS Check and Map

Words that describe the center

k-means splits Rennes into 7 areas. The center, cluster_4, is the real visitor place: 1,080 photos from 135 people. The north-east is the La Rennaise event at one point, and the south is a few local photographers rather than a tourist crowd.

One thing I found that the center has 135 photographers, but its strongest text pattern is one local photographer’s signature, so tag frequency alone does not name a tourist site which is our goal.The search did what it should: it found the words that appear together most often. In this dataset, those words belong to a photographer, not to a site. Check the last Sorter Result.

Places Named by other Visitors

The atana photographer is removed, then the same word search is run again. The place that remains is Parc du Thabor, in 32 photos. The other frequent groups are architecture and Instagram upload tags.

Hierarchical Clustering

k-Means is the useful split: seven areas, including the center and the La Rennaise race. Hierarchical clustering puts about 90% of the photos in two groups and spends three groups on isolated points of 8, 4, and 1 photo.

Reporting

Data Input

1

2

3

4

5

6

7

8

9

10

Author
Muhammad Waseem JAVED
MS Data Science, University of Rennes, France

Supervision: Prof. Bertrand COUASNON

The 30 words that appear most often in center photos.
Bar Chart
Frequent word groups in the center. Support is the number of photos that contain the whole group. I use Apriori Algorithm
Item Set Finder (Borgelt)
Each row ID appears once. It is a line number, so it cannot reveal a repeated photo
GroupBy
Put the most common word groups first. The top groups appear in about 31% of the center photos.
Sorter
One 0/1 column per word for each center photo. 1,042 photos kept a usable text.
Document Vector
Pack those word columns into one bit vector so the frequent word groups can be searched.
Create Bit Vector
3,990 photos appear once, 5 appear twice, and 200 photos are copied 254 times.
Bar Chart
Snowball Stemmer
Duplicate check while row ID is still present: 54,800 rows remain, so nothing is removed.
Duplicate Row Filter
Bag Of Words Creator
One row per photo. The file contains 4,195 distinct photos.
GroupBy
Stop Word Filter
Repetition summary: 3,990 photos appear once, 5 appear twice, and 200 photos are copied 254 times.
GroupBy
Remove the photographer signature and the generic location tags. Place words stay.
CSV Reader
Checking on text_status what is missing/present
GroupBy
Text on the 4,195 photos: 2,072 have tags and a title, 1,997 have only a title, 8 have only tags, and 118 have no text at all
Bar Chart
On the 4,195 unique photos, 2,115 have no tags and 126 have no title
Statistics
Classify each photo by whether its tags and title are missing.
Rule Engine
Create Bit Vector
Item Set Finder (Borgelt)
Document Vector
Replace missing tags and titles with an empty string so they can be joined. No placeholder word is added.
Missing Value
Keep the single words only. Pairs such as parc + thabor repeat a word that is already counted on its own.
Row Filter
Combine tags and title into one text field for the later description of each place.
String Manipulation
Keep the 15 most frequent words. The table is already sorted by support.
Top k Row Filter
After removing the atana photographer, the strongest place group is Parc du Thabor (32 photos). The other frequent groups are architecture and Instagram upload tags.
Sorter
GPS consistency check for Rennes. All 4,195 photos fall inside the city box, so none need to remove.
Rule-based Row Filter
Leave out the atana studio photographer so the search uses the other visitors’ photos.
Row Filter
One marker per location. Photos with identical coordinates are counted together so the map can draw them
GroupBy
Hierarchical Clustering
Sorting Photos based on their occurance
Sorter
Keep the location, the photographer, the text, and the date. Drop the redundant date parts.
Column Filter
ItemSet is a list, so the cell is [thabor]. This turns it into the plain word the bar chart can use.
Ungroup
Cleaned photos on the map of Rennes
OSM Map View
Main words in the center. Architecture, street names, and Parc du Thabor lead. Night photos and the Vilaine follow.
Bar Chart
Group the 4,195 photos by location. First try with 7 clusters.
k-Means
Largest location: 1,560 photos at one coordinate, titled La Rennaise 2017. The next places are far smaller.
Top k Row Filter
Top 10 locations. One coordinate holds 1,560 photos (La Rennaise 2017). Every other place has far fewer.
Bar Chart
Raw Flickr file of Rennes: 54,800 rows
CSV Reader
Exclude row ID. It is a line number, so it hides duplicate photos
Column Filter
Keep the first copy of each photo. 4,195 unique photos remain
Duplicate Row Filter
OSM Map View
Bar Chart
GroupBy
Color Palette Designer
One color per cluster, so each candidate point of interest is visible on the map.
OSM Map View
Saves the center words with their photo counts. Same rows as the bar chart.
CSV Writer
Photos that share the same coordinates are counted together so the map can draw them
GroupBy
Drops the upload tags, the regional tags, and the date. The file keeps the place and theme words.
Rule-based Row Filter
Column Filter
The 6 clusters on the map of Rennes. Each color is a candidate point of interest.
Color Palette Designer
Column Renamer
Center photos that still have text.
Rule-based Row Filter
Each center photo becomes a document. The words come from the tags and the title, and the author is the photographer.
Strings to Document
Seven areas of Rennes. The center (cluster_4) is shared by 135 photographers. The north-east (cluster_1) is the La Rennaise event. The south (cluster_2) is a few local photographers.
GroupBy
GroupBy
Photos per cluster. cluster_1 (1,664), cluster_4 (1,080) and cluster_2 (873) hold most of the photos. cluster_4 is the center, shared by 135 photographers.
Bar Chart
Table Creator
Remove punctuation from the center photos.
Punctuation Erasure
Drop words shorter than 3 letters.
N Chars Filter
Joiner
Remove numbers from the text.
Number Filter
Column Renamer
One row per k-Means area: how many photos, how many photographers, what the area is, and what to do with it.
CSV Writer
Remove English empty words such as the, and, and of.
Stop Word Filter
Reduce each word to its root.
Snowball Stemmer
Remove French empty words such as le, la, de, and et
Stop Word Filter
How many center photos contain each word.
GroupBy
Sort in Desc
Sorter
One row per word in each photo.
Bag Of Words Creator
Lowercase the title and the text before the document is created, so Rennes and rennes become one word.
Case Converter
Top 30
Top k Row Filter

Nodes

Extensions

Links

Download

To use this workflow in KNIME, download it from the below URL and open it in KNIME:

Download Workflow
Created by: ardjood
Created at: 2026-10-08
On NodePit since: 2026-10-18
Last update: 2026-10-18
Created with KNIME version: v5.12.0

Deploy, schedule, execute, and monitor your KNIME workflows locally, in the cloud or on-premises – with our brand new NodePit Runner.

Try NodePit Runner!