The document discusses the creation of a Converso chatbot platform by a team including Talha Arshad Khan and others. It explains various algorithms for k-nearest neighbor search, highlighting the brute force method's computation of distances and comparisons with inverted file indexes and hierarchical navigable small world (HNSW) graphs. The content also includes pseudocode for the brute force approach and comparisons of algorithm efficiency and complexities.

1. CS-09
xx / xx / 2024
CONVERSO
Chatbot Creation Platform

2. Talha Arshad Khan
21B-206-Cs
Syed Usman Riaz
21B-218-Cs
Shayan Naushad
21B-047-Cs
Anas Khan
21B-111-Cs
Laraib Jamal
21B-091-Cs
Meet The Team

3. POC
01
Cost
Domain Specific
Data Source
02
03

4. Actor Use Case
Diagram

5. Coding Architecture

6. Class Diagram
+ public
# protected
- private

7. Front Ends

8. Sequence
Diagram

9. ERD
Diagram

10. Object Diagram

11. Algorithms
Significance
Brute Force KNN Search
The brute force approach computes similarity distances from each query point to all the
reference points using a predefined metric. In our case, we are using IndexFlat L2 distance
metric that has the formula:
d(q,v) = √ (∑ (qi−vi)^2 )
It is versatile and widely applicable in domain such as information retrieval based on similarity.
It provides solutions for small datasets and is easily adaptable to various metrics.

12. Pseudocode
Input:
Q: a set of query points.
R: a set of reference points.
Output:
A list of ‘k’ reference points for each query point.
Steps:
1. For each query point q Q do:
∈
1) Compute distances between q and all r R.
∈
2) Sort the computed distances.
3) Select k-nearest reference points corresponding to k smallest distances.

13. Similar Algorithms
Brute Force KNN vs. Inverted File Index:
Algorithm Speed Accuracy Time Complexity Space Complexity
Brute-Force KNN Slow for large
datasets
Exact O(n.m.d) [1] O(m.d)
Inverted File Index
(IVF)
Faster (coarse
clusters)
Approximate O(W.D.DL) [2] O(W.D) [2]
[1] (IEEE)K-NEAREST NEIGHBOR SEARCH FAST GPU-BASED IMPLEMENTATIONS
[2] (IAENG InternationalJournalCS)Inverted Index Construction Algorithms For large dS

14. Algorithm Speed Accuracy Time Complexity Space Complexity
Brute-Force
KNN
Slow for large
datasets
Exact O(n.m.d) [1] O(m.d)
HNSW Very fast (graph
traversal)
High recall Construction:
O(n.log(n))
Query: O(log(n)) [2]
O(Mmax0
+mL.Mmax).bytes_per_link
[2]
Similar Algorithms
Brute Force KNN vs. Hierarchical Navigable Small World Graphs (HNSW):
[1] (IEEE)K-NEAREST NEIGHBOR SEARCH FAST GPU-BASED IMPLEMENTATIONS
[2] (RG)Efficient and robust approximate NN search

15. Similar Algorithms Big O Notation:
HNSW vs. Inverted File Index:
Algorithm Best Case Average Case Worst Case
Inverted File Index(IVF) Ω(D) Θ(D) O(W.D.DL) [3]
Hierarchical Navigable
Small World (HNSW)
Ω(logn) Θ(logn) Construction:
O(n.log(n))
Query: O(log(n)) [4]
BF KNN Search Ω(n.m.d) Θ(n.m.d) O(n.m.d)
[1]Inverted Index
[2]Understanding Recall in HNSW Search
[3] (IAENG InternationalJournalCS)Inverted Index Construction Algorithms For large dS
[4] (RG)Efficient and robust approximate NN search

16. Plagiarism Report

17. Thank You