implicit-and-explicit-aspects-Extraction-in-Restaurant-Reviews-Domain

Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->

Extracting Implicit and Explicit Aspects from Restaurant Reviews using RoBERTa-Large Variant with Benchmark Efficiency and Custom Dataset We present a groundbreaking approach to extracting implicit and explicit aspects from restaurant reviews in the domain. Leveraging the powerful RoBERTa-Large variant, our method achieves remarkable performance while utilizing a custom dataset. Our research addresses the challenging task of aspect extraction, which involves identifying both explicit aspects explicitly mentioned in reviews, as well as implicit aspects that are indirectly referred to. By employing RoBERTa-Large, a state-of-the-art language model, we leverage its advanced contextual understanding to capture nuanced information from textual data. To ensure the efficiency and accuracy of our approach, we benchmarked our system against existing methods in the field. The results were outstanding, highlighting the superiority of our approach in terms of precision, recall, and overall performance. Furthermore, we developed a custom dataset tailored specifically to the restaurant domain, encompassing a diverse range of reviews from various platforms. This dataset allowed us to train our model with domain-specific knowledge, leading to better aspect extraction outcomes.

Overall, our research presents a novel and efficient solution for aspect extraction from restaurant reviews. By employing the RoBERTa-Large variant and a carefully curated custom dataset, we have achieved remarkable results that surpass existing approaches. This breakthrough has significant implications for sentiment analysis, opinion mining, and other natural language processing applications in the restaurant domain.

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

• Developed by: Ali Haider
• Shared by: Ali Haider
• Model type: Bert Varinet
• Language(s) (NLP): English (Restaurant Domain Reviews)
• Finetuned from model: Roberta Large

Uses

Aspects Extraction Model in Restaurant Domain aimns to extract the Implicit and explicit aspects that might be speifified in the Reviews we can use our model for vairous purposes such as

1. Aspects extraction from the reviews Sentences and classification under 34 aspects-categoires.
2. Aspects based Restaurant Recommendation system
3. Restaurant Reviews Analysis

Out-of-Scope Use

Model has been tuned to classifiy the out of scope sentences into the General.

How to Get Started with the Model

Sample Sentence: The food was very delicious, elegant Ambience and Decoration , floors were clean and most importantly the food was affoardable. Expected Output:

Food-Taste Food-Price Restaurant-Decoration Restaurant-Atmosphere Restaurant-Hygiene

Training Details

Roberta-large varient is used with 10678 data entires each of the sentence is classified under serveral Aspects they might belong to and trained till the Validation loss not improving till 3 epochs.

Training Data

Reviews are tokenized into sentences and 10678 unique sentences are annotated for training. Aspects are Categorized under 4 categories

    Restaurants (Restaurants and Ambience Merged)
    	Atmopshere
    	Building
    	Location
    	Features
    	Hygiene
    	Kitchen
    	Recommendation
    	View
    	Decoration
    	Seating Plan
    	Options
    	Experience
    	General
     
    Service (Staff and Service Merged)
    	Behavior
    	Wait Time
    	General
    	Experience
     
    Food (Food and Drinks Merged)
    	Cuisine
    	Deals
    	Diet Options
    	Ingredients
    	Menu
    	Kitchen
    	Portion
    	Presentation
    	Price
    	Quality
    	Taste
    	Flavor
    	Recommendation
    	Experience
    	Dishes
    	General
     
    General (Out of Domain and Contextless Sentences)
    	General

Training Hyperparameters

lr=2e-5 eps=1e-8 batch_size=32

Evaluation and Results

Classification Report precision recall f1-score support

             FOOD-CUISINE       0.69      0.83      0.76        65
               FOOD-DEALS       0.81      0.75      0.78        40
         FOOD-DIET_OPTION       0.73      0.93      0.82        71
          FOOD-EXPERIENCE       0.38      0.44      0.40        55
              FOOD-FLAVOR       0.83      0.94      0.88        63
             FOOD-GENERAL       0.65      0.78      0.71       141
          FOOD-INGREDIENT       0.77      0.80      0.78        54
             FOOD-KITCHEN       0.50      0.60      0.55        35
                FOOD-MEAL       0.72      0.74      0.73       208
                FOOD-MENU       0.80      0.89      0.84       136
             FOOD-PORTION       0.90      0.91      0.90        76
        FOOD-PRESENTATION       0.82      0.94      0.87        33
               FOOD-PRICE       0.74      0.88      0.80        57
             FOOD-QUALITY       0.61      0.66      0.63       102
      FOOD-RECOMMENDATION       0.65      0.47      0.55        32
               FOOD-TASTE       0.79      0.84      0.82       114
          GENERAL-GENERAL       0.98      0.88      0.93       163
    RESTAURANT-ATMOSPHERE       0.73      0.79      0.76       170
      RESTAURANT-BUILDING       0.90      0.86      0.88        44
    RESTAURANT-DECORATION       0.95      0.84      0.89        44
    RESTAURANT-EXPERIENCE       0.67      0.60      0.63       189
      RESTAURANT-FEATURES       0.55      0.76      0.64        75
       RESTAURANT-GENERAL       0.45      0.49      0.47        47
       RESTAURANT-HYGIENE       0.94      0.92      0.93        51
       RESTAURANT-KITCHEN       0.82      0.85      0.84        33
      RESTAURANT-LOCATION       0.59      0.78      0.67        69
       RESTAURANT-OPTIONS       0.42      0.41      0.41        32
RESTAURANT-RECOMMENDATION       0.62      0.71      0.67        49
  RESTAURANT-SEATING_PLAN       0.78      0.82      0.80        68
          RESTAURANT-VIEW       0.80      0.88      0.84        42
        SERVICE-BEHAVIOUR       0.65      0.87      0.74       127
       SERVICE-EXPERIENCE       0.31      0.24      0.27        21
          SERVICE-GENERAL       0.74      0.81      0.77       162
        SERVICE-WAIT_TIME       0.86      0.85      0.86        94

                micro avg       0.72      0.78      0.75      2762
                macro avg       0.71      0.76      0.73      2762
             weighted avg       0.73      0.78      0.75      2762
              samples avg       0.75      0.78      0.75      2762
          

Accuracy 0.9801993831240361

Confusin Matrix [[[2047, 24], [ 11, 54]],

     [[2089,    7],
      [  10,   30]],

     [[2041,   24],
      [   5,   66]],

     [[2041,   40],
      [  31,   24]],

     [[2061,   12],
      [   4,   59]],

     [[1936,   59],
      [  31,  110]],

     [[2069,   13],
      [  11,   43]],

     [[2080,   21],
      [  14,   21]],

     [[1869,   59],
      [  55,  153]],

     [[1969,   31],
      [  15,  121]],

     [[2052,    8],
      [   7,   69]],

     [[2096,    7],
      [   2,   31]],

     [[2061,   18],
      [   7,   50]],

     [[1991,   43],
      [  35,   67]],

     [[2096,    8],
      [  17,   15]],

     [[1997,   25],
      [  18,   96]],

     [[1970,    3],
      [  19,  144]],

     [[1917,   49],
      [  36,  134]],

     [[2088,    4],
      [   6,   38]],

     [[2090,    2],
      [   7,   37]],

     [[1890,   57],
      [  75,  114]],

     [[2015,   46],
      [  18,   57]],

     [[2061,   28],
      [  24,   23]],

     [[2082,    3],
      [   4,   47]],

     [[2097,    6],
      [   5,   28]],

     [[2029,   38],
      [  15,   54]],

     [[2086,   18],
      [  19,   13]],

     [[2066,   21],
      [  14,   35]],

     [[2052,   16],
      [  12,   56]],

     [[2085,    9],
      [   5,   37]],

     [[1950,   59],
      [  17,  110]],

     [[2104,   11],
      [  16,    5]],

     [[1927,   47],
      [  30,  132]],

     [[2029,   13],
      [  14,   80]]]

Average Validation loss 0.06330019191129883

Model Card Authors

Ali Haider

Model Card Contact

+923068983139 [email protected]