TaxoSplitter: Business Router (e5-base)

This is a fine-tuned version of intfloat/e5-base-v2 designed specifically to route raw Point of Interest (POI) and business categories (e.g., from Google Maps, Yelp, or CRM data) into 14 strict B2B sales verticals.

It solves the "Suffix Poisoning" problem inherent in zero-shot embedding models, where generic suffixes (like "station" or "company") overpower the semantic meaning of the core noun.

This is a sentence-transformers model finetuned from intfloat/e5-base-v2. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, classification, clustering, and more.

Model Details

• Base Model: intfloat/e5-base-v2
• Architecture: Sentence Transformer / Feature Extraction
• Dimensions: 768
• Pooling: Mean Pooling
• Training Objective: Contrastive Learning (CosineSimilarityLoss)
• Training Data: 30,345 positive and negative pairs generated from a 120B LLM-grounded dataset.

The 14 Domains

The model is trained to route categories into the following distinct clusters: Healthcare, PetCare, Automotive, Construction, Cleaning, RealEstate, FoodDining, Lodging, BeautyWellness, FitnessSports, Retail, B2BCorporate, Manufacturing, PublicNonProfit, and ARCHIVE.

Usage (Sentence-Transformers)

pip install -U sentence-transformers


### Model Description
- **Model Type:** Sentence Transformer
- **Base model:** [intfloat/e5-base-v2](https://huggingface.co/intfloat/e5-base-v2) <!-- at revision f52bf8ec8c7124536f0efb74aca902b2995e5bcd -->
- **Maximum Sequence Length:** 512 tokens
- **Output Dimensionality:** 768 dimensions
- **Similarity Function:** Cosine Similarity
- **Supported Modality:** Text
<!-- - **Training Dataset:** Unknown -->
<!-- - **Language:** Unknown -->
<!-- - **License:** Unknown -->

### Model Sources

- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)
- **Repository:** [Sentence Transformers on GitHub](https://github.com/huggingface/sentence-transformers)
- **Hugging Face:** [Sentence Transformers on Hugging Face](https://huggingface.co/models?library=sentence-transformers)

### Full Model Architecture

SentenceTransformer( (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'}) (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'mean', 'include_prompt': True}) (2): Normalize({}) )

## Usage

### Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

```bash
pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("MajidFQ/taxoSplitter_biz-router-e5")
# Run inference
sentences = [
    'veterinarian',
    'RULE: The business sells specialized professional knowledge, legal,\n        financial, technology, or administrative services — primarily to\n        other businesses or sophisticated clients. The deliverable is advice,\n        a document, software, or a managed service, not a physical product\n        or in-person experience.\n\n        INCLUDES: Law firms, attorneys, legal services, notary publics,\n        paralegals. Accounting firms, accountants, auditors, tax preparation,\n        payroll services, bookkeepers. Financial planners, financial advisors,\n        investment services, insurance agencies, banks, credit unions, debt\n        collectors. Management consultants, business consultants, HR\n        consultants. Marketing agencies, advertising agencies, branding\n        agencies, PR firms, SEO agencies, digital agencies. Software\n        companies, IT service companies, managed IT providers, cybersecurity\n        firms, web hosting, ISPs, data centers, cloud services, tech support.\n        Staffing agencies, employment agencies, executive recruiters.\n        Architects, engineering consultants (office-based), design firms.\n        Private investigators, process servers, court reporters, notaries.\n        Commercial printing, sign companies (design and print).\n\n        DOES NOT INCLUDE: On-site trade contractors (D04_Construction).\n        Government agencies (D14_PublicNonProfit). Hospitals and medical\n        practices (D01_Healthcare). Software retail stores (D11_Retail).',
    'RULE: The business provides health, grooming, boarding, or care\n        services where the primary patient or client is an animal.\n\n        INCLUDES: Veterinarians, animal hospitals, veterinary specialists,\n        emergency animal clinics, pet dental care. Dog groomers, cat groomers,\n        mobile pet groomers. Dog training, pet obedience schools. Boarding\n        kennels, doggy day care, cat boarding, pet sitting services, dog\n        walking services. Aquarium maintenance, exotic animal care,\n        livestock veterinarians. Pet cremation, animal shelters and rescues.\n\n        DOES NOT INCLUDE: Pet supply stores and pet food retail (D11_Retail).\n        Animal farms raising livestock for food production (D13_Manufacturing).\n        Human healthcare (D01_Healthcare).',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000, -0.0284,  0.9810],
#         [-0.0284,  1.0000, -0.0252],
#         [ 0.9810, -0.0252,  1.0000]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 30,345 training samples
• Columns: sentence_0, sentence_1, and label
• Approximate statistics based on the first 100 samples:

  	sentence_0	sentence_1	label
  type	string	string	float
  modality	text	text
  details	min: 3 tokensmean: 4.69 tokensmax: 8 tokens	min: 151 tokensmean: 235.07 tokensmax: 286 tokens	min: 0.0mean: 0.07max: 1.0

• Samples:

  sentence_0	sentence_1	label
  building materials market	RULE: The business provides health, grooming, boarding, or care services where the primary patient or client is an animal. INCLUDES: Veterinarians, animal hospitals, veterinary specialists, emergency animal clinics, pet dental care. Dog groomers, cat groomers, mobile pet groomers. Dog training, pet obedience schools. Boarding kennels, doggy day care, cat boarding, pet sitting services, dog walking services. Aquarium maintenance, exotic animal care, livestock veterinarians. Pet cremation, animal shelters and rescues. DOES NOT INCLUDE: Pet supply stores and pet food retail (D11_Retail). Animal farms raising livestock for food production (D13_Manufacturing). Human healthcare (D01_Healthcare).	0.0
  meditation center	RULE: The entry is not a business — it is a geographic location, natural feature, public landmark, or abstract term with no identifiable commercial entity that a B2B sales team could contact. INCLUDES: Natural features — beaches, rivers, lakes, mountains, waterfalls, caves, forests, bays, islands, peninsulas, cliffs, springs, inlets, ridges. Public parks — national parks, state parks, city parks, nature preserves, wildlife refuges, botanical gardens. Public infrastructure as places — bridges, dams, canals, highways, train stations (the physical location, not a transit company). Purely geographic or civic landmarks — plazas, monuments, scenic overlooks, historical sites as places (not as museum businesses). Abstract or overly generic terms with no specific commercial entity: "services", "industry", "company", "store", "center" alone. DOES NOT INCLUDE: Businesses operating at natural...	0.0
  gun club	RULE: The business provides overnight accommodation or organized travel and tourism experiences. INCLUDES: Hotels, motels, boutique hotels, luxury resorts, beach resorts, mountain resorts, casino resorts, extended-stay hotels, bed and breakfasts, hostels, guest houses, inns, lodges, serviced apartments, vacation rental operators, RV parks, campgrounds, glamping sites, cabin rentals. Tour operators, travel agencies, cruise agencies, sightseeing tour companies, boat tour operators, helicopter tour operators, hot air balloon rides, guided hiking tours, adventure tourism operators. Airport shuttles and travel transport services. DOES NOT INCLUDE: Restaurants and bars (D07_FoodDining). Real estate managing long-term rentals (D06_RealEstate). Amusement and recreation venues (D10_FitnessSports).	0.0

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss",
      "cos_score_transformation": "torch.nn.modules.linear.Identity"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• num_train_epochs: 4
• per_device_eval_batch_size: 16
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

• per_device_train_batch_size: 16
• num_train_epochs: 4
• max_steps: -1
• learning_rate: 5e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 1
• average_tokens_across_devices: True
• max_grad_norm: 1
• label_smoothing_factor: 0.0
• bf16: False
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 16
• prediction_loss_only: True
• eval_on_start: False
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: None
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: False
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: False
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• dataloader_multiprocessing_context: None
• dataloader_in_order: True
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_static_graph: None
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: None
• fsdp_config: None
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• local_rank: -1
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}
• warmup_ratio: None

Training Logs

Training Time

• Training: 2.1 hours

Framework Versions

• Python: 3.13.15
• Sentence Transformers: 5.7.0
• Transformers: 5.16.1
• PyTorch: 2.11.0+cu128
• Accelerate: 1.14.0
• Datasets: 4.0.0
• Tokenizers: 0.23.1

Additional Resources

• Training and Finetuning Embedding Models with Sentence Transformers: the end-to-end guide for training or finetuning Sentence Transformer models.
• Introduction to Matryoshka Embedding Models: variable-size embeddings that can be truncated with minimal quality loss.
• Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval: post-training compression of embedding vectors.
• Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video models through the same API.
• Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: train multimodal embedding models, with a Visual Document Retrieval walkthrough.

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}