Change the word level to char level, I get acc 0

[hertz-pj]

No description provided.

[hertz-pj]

When I use your model to train on the TREC datasets, I found something interesting. If I change the your read_TREC function like this：

def read_TREC():
    data = {}

    def read(mode):
        x, y = [], []

        with open("../data/TREC/TREC_" + mode + ".txt", "r", encoding="utf-8") as f:
            for line in f:
                if line[-1] == "\n":
                    line = line[:-1]
                y.append(line.split(":")[0])
                x.append(line.split(":")[1])

        x, y = shuffle(x, y)

        if mode == "train":
            dev_idx = len(x) // 10
            data["dev_x"], data["dev_y"] = x[:dev_idx], y[:dev_idx]
            data["train_x"], data["train_y"] = x[dev_idx:], y[dev_idx:]
        else:
            data["test_x"], data["test_y"] = x, y

    read("train")
    read("test")

    return data

I get the very high Accuracy 0.97+

[zhangmingfang123]

I want to classify Chinese sentences with about 50 words. Is this model effective?

[zhangmingfang123]

As for the question I raised above, do you have any suggestions for model changes?