app.py

import gradio as gr
from gensim.models import KeyedVectors


def isNoneWords(word):
    if word is None or len(word)==0 or word not in model.key_to_index:
        return True
    else:
        return False
    
def word_analogy(word1, word2, word3):
    analogy_words = model.similar_by_vector(model.word_vec(word1) - model.word_vec(word2) + model.word_vec(word3))
    sim_res = ""
    for item in analogy_words:
        sim_res += f'{item[0]}: {round(item[1], 4)}\n'
    return sim_res
    
def similarity_route(word1, word2):
    if isNoneWords(word1) or isNoneWords(word2):
        return "word is null or not in model!"
    else:
        return float(model.similarity(word1, word2))


def top_similarity_route(word):
    if isNoneWords(word):
        return "word is null or not in model!"
    else:
        top_similar_words = model.similar_by_word(word, topn=20, restrict_vocab=None)
        sim_res = ""
        for item in top_similar_words:
            sim_res += f'{item[0]}: {round(item[1], 4)}\n'
        return sim_res

def top_similar_words_layout():
    with gr.Column():
        with gr.Row():
            with gr.Column():
                word = gr.Textbox(lines=1, label='Input word', placeholder='Input word here')
                with gr.Row():
                    clear = gr.ClearButton()
                    submit = gr.Button("Submit")
            output = gr.Textbox(lines=20, label='Similar words', placeholder='Output here')

    submit.click(fn=top_similarity_route, inputs=[word], outputs=[output])
    
    examples=[['兔子', '松鼠']]
    ex = gr.Examples(
        examples=examples,
        fn=top_similarity_route,
        inputs=[word],
        outputs=[output],
        cache_examples=False, 
        run_on_click=False
    )
    
    
def similarity_layout():
    with gr.Column():
        with gr.Row():
            with gr.Column():
                with gr.Row():
                    word1 = gr.Textbox(lines=1, label='Input word1', placeholder='Input word1 here')
                    word2 = gr.Textbox(lines=1, label='Input word2', placeholder='Input word2 here')
                with gr.Row():
                    clear = gr.ClearButton()
                    submit = gr.Button("Submit")
            output = gr.Textbox(lines=1, label='Similar words', placeholder='Output here')

    submit.click(fn=similarity_route, inputs=[word1, word2], outputs=[output])
    
    examples=[['淘宝', '京东', 0.7887385]]
    ex = gr.Examples(
        examples=examples,
        fn=similarity_route,
        inputs=[word1, word2],
        outputs=[output],
        cache_examples=False, 
        run_on_click=False
    )
    
def word_analogy_layout():
    with gr.Column():
        with gr.Row():
            with gr.Column():
                with gr.Row():
                    word1 = gr.Textbox(lines=1, label='Input word1', placeholder='Input word1 here')
                    word2 = gr.Textbox(lines=1, label='Input word2', placeholder='Input word2 here')
                    word3 = gr.Textbox(lines=1, label='Input word3', placeholder='Input word3 here')
                with gr.Row():
                    clear = gr.ClearButton()
                    submit = gr.Button("Submit")
            output = gr.Textbox(lines=1, label='Analogy words', placeholder='Output here')

    submit.click(fn=word_analogy, inputs=[word1, word2, word3], outputs=[output])
    
    examples=[['国王', '男人', '女人', '王后']]
    ex = gr.Examples(
        examples=examples,
        fn=word_analogy,
        inputs=[word1, word2, word3],
        outputs=[output],
        cache_examples=False, 
        run_on_click=False
    )

if __name__ == '__main__':
    model = KeyedVectors.load_word2vec_format('tencent-ailab-embedding-zh-d100-v0.2.0-s.txt', binary=False)
    title = 'Calculate word similarity based on Tencent AI Lab Embedding'

    with gr.Blocks() as demo:
        gr.HTML(title)
        with gr.Column(elem_id="col-container"):
            with gr.Tab("Top similar words"):
                top_similar_words_layout()
            with gr.Tab("Similarity of words"):
                similarity_layout()
            with gr.Tab("Word analogy"):
                word_analogy_layout()
            
    demo.queue(max_size=64).launch()