Skip to content

Usage Examples

This page provides practical examples of using AniSearch Model for various scenarios.

Basic Search Examples

Finding Action Anime with a Specific Theme

python src/main.py search --type anime --query "An action anime about a protagonist who gains supernatural powers and fights evil organizations"

Top 5 matches for 'An action anime about a protagonist who gains supernatural powers and fights evil organizations':

1. My Hero Academia (Score: 0.87)
   ID: 31964
   Synopsis excerpt: Izuku Midoriya, a boy born without superpowers in a world where they have become commonplace, but who still dreams of becoming a superhero himself, is scouted by the world's greatest hero who shares his powers with Izuku after recognizing his potential...

2. One Punch Man (Score: 0.82)
   ID: 30276
   Synopsis excerpt: The seemingly ordinary and unimpressive Saitama has a rather unique hobby: being a hero. In order to pursue his childhood dream, he trained relentlessly for three years—and lost all of his hair in the process...

...

Finding Romance Manga

python src/main.py search --type manga --query "A high school romance manga where the main characters start off disliking each other"

Top 5 matches for 'A high school romance manga where the main characters start off disliking each other':

1. Kaguya-sama: Love is War (Score: 0.91)
   ID: 90125
   Synopsis excerpt: Considered a genius due to having the highest grades in the country, Miyuki Shirogane leads the prestigious Shuchiin Academy's student council as its president, working alongside the beautiful and wealthy vice president Kaguya Shinomiya. The two are often regarded as the perfect couple...

...

Interactive Mode Examples

Anime Search Session

$ python src/main.py search --type anime --interactive

=== AniSearch Model Interactive Mode ===
Type a description to search for anime (Ctrl+C to exit):

> Anime with time travel

Top 5 matches for 'Anime with time travel':
1. Steins;Gate (Score: 0.89)
2. Erased (Score: 0.85)
3. Re:Zero -Starting Life in Another World- (Score: 0.79)
4. Tokyo Revengers (Score: 0.78)
5. The Girl Who Leapt Through Time (Score: 0.77)

Enter another query:

> Psychological thriller with mind games

Top 5 matches for 'Psychological thriller with mind games':
1. Death Note (Score: 0.93)
2. Code Geass (Score: 0.86)
3. Monster (Score: 0.85)
4. Psycho-Pass (Score: 0.82)
5. The Promised Neverland (Score: 0.79)

Advanced Usage Examples

Using Different Models

python src/main.py search --type anime --query "Space western with bounty hunters" --model "cross-encoder/ms-marco-TinyBERT-L-2"

python src/main.py search --type anime --query "Space western with bounty hunters" --model "cross-encoder/ms-marco-MiniLM-L-6-v2"

python src/main.py search --type anime --query "Space western with bounty hunters" --model "cross-encoder/ms-marco-MiniLM-L-12-v2"

python src/main.py search --type manga --query "Isekai adventure with game mechanics"

python src/main.py search --type manga --query "Isekai adventure with game mechanics" --include-light-novels

Custom Training Examples

Fine-tuning on Anime Data

python src/main.py train --type anime --model "cross-encoder/ms-marco-MiniLM-L-6-v2" \
    --epochs 3 --batch-size 16 --learning-rate 2e-5 --max-samples 10000

Creating training dataset...
Generated 10000 training examples
Starting training with batch size 16 for 3 epochs...
Epoch 1/3: 100%|██████████| 625/625 [04:12<00:00, 2.48it/s, loss=0.142]
Evaluation: 100%|██████████| 32/32 [00:13<00:00, 2.35it/s]
Epoch 2/3: 100%|██████████| 625/625 [04:10<00:00, 2.49it/s, loss=0.079]
Evaluation: 100%|██████████| 32/32 [00:13<00:00, 2.37it/s]
Epoch 3/3: 100%|██████████| 625/625 [04:11<00:00, 2.48it/s, loss=0.055]
Evaluation: 100%|██████████| 32/32 [00:13<00:00, 2.36it/s]

==================================================
Training completed successfully!
Fine-tuned model saved to: model/fine-tuned/anime-model-2023-10-25-12-30-45
To use this model for search:
  python src/main.py search --type anime --model "model/fine-tuned/anime-model-2023-10-25-12-30-45" --query "Your query"
==================================================

Creating Labeled Data Without Training

python src/main.py train --type manga --create-labeled-data "data/labeled_manga.csv"

Creating labeled dataset...
Generated 50000 labeled examples
Labeled data created and saved to: data/labeled_manga.csv

Query Variation Process

The search system improves results by generating variations of the original query:

flowchart LR
    A[Base Queries] --> B[Create Empty Variations List]
    B --> C[Add Original Queries]
    C --> D[Select Random Templates]
    D --> E[Format Templates with Queries]
    E --> F[Return Combined Variations]

    style A fill:#e1f5fe,stroke:#0288d1
    style F fill:#e8f5e9,stroke:#4caf50
Press "Alt" / "Option" to enable Pan & Zoom

For example, a simple query like "time travel" might be expanded to:

  • time travel
  • Looking for anime about time travel
  • I want to watch anime with time travel
  • Find me anime where time travel

This makes the search more robust to different ways of expressing the same information need.