app.py

from dash import Dash, dcc, html, Input, Output, State
import dash.dependencies as dd
import plotly.express as px
import pandas as pd

# Load DataFrame from CSV
df_result_bad_distil_2 = df = pd.read_csv(
    "hf://datasets/Jbddai/customer_reviews/bad_distil_2_with_cluster_labels_cleaned_company.csv"
)
df_result_good_distil_2 = df = pd.read_csv(
    "hf://datasets/Jbddai/customer_reviews/good_distil_2_with_cluster_labels_cleaned_company.csv"
)


def preprocess_data_for_slider_marks(df):
    min_label = df["labels"].min()
    max_label = df["labels"].max()
    min_cluster_rank = df["cluster_rank"].min()
    max_cluster_rank = df["cluster_rank"].max()

    label_marks = {i: str(i + 1) for i in range(min_label, max_label + 1, 10)}
    cluster_rank_marks = {i: str(i + 1) for i in range(min_cluster_rank, max_cluster_rank + 1, 10)}

    return label_marks, min_label, max_label, cluster_rank_marks, min_cluster_rank, max_cluster_rank


label_marks, min_label, max_label, cluster_rank_marks, min_cluster_rank, max_cluster_rank = (
    preprocess_data_for_slider_marks(df_result_good_distil_2)
)

sentiment_options = [
    {"label": "gut", "value": "gut"},
    {"label": "schlecht", "value": "schlecht"},
]

app = Dash(__name__)

app.layout = html.Div(
    [
        html.Div(
            [
                html.H4("Interactive Plot of Customer Reviews"),
                html.Div(
                    [
                        html.P("Select sentiment:"),
                        dcc.Dropdown(
                            id="sentiment-dropdown",
                            options=sentiment_options,
                            value="gut",
                            style={"width": "50%", "margin": "auto"},
                            clearable=False,
                            multi=False,
                        ),
                    ],
                    style={"width": "50%", "margin": "auto"},
                ),
                html.Div(
                    [
                        html.P("Select range of labels:"),
                        dcc.RangeSlider(
                            id="label-range-slider",
                            min=df_result_good_distil_2["labels"].min(),
                            max=df_result_good_distil_2["labels"].max(),
                            step=1,
                            value=[
                                df_result_good_distil_2["labels"].min(),
                                df_result_good_distil_2["labels"].max(),
                            ],
                            marks={
                                i: str(i + 1)
                                for i in range(
                                    df_result_good_distil_2["labels"].min(),
                                    df_result_good_distil_2["labels"].max() + 1,
                                    10,
                                )
                            },
                            tooltip={"always_visible": True, "placement": "bottom"},
                        ),
                        html.Button("Reset Range", id="reset-button", n_clicks=0),
                    ],
                    style={"width": "50%", "margin": "auto"},
                ),
                html.Div(
                    [
                        html.P("Select range of cluster by rank/popularity (by number of reviews descending):"),
                        dcc.RangeSlider(
                            id="cluster-rank-slider",
                            min=df_result_good_distil_2["cluster_rank"].min(),
                            max=df_result_good_distil_2["cluster_rank"].max(),
                            step=1,
                            value=[
                                df_result_good_distil_2["cluster_rank"].min(),
                                df_result_good_distil_2["cluster_rank"].max(),
                            ],
                            marks={
                                i: str(i + 1)
                                for i in range(
                                    df_result_good_distil_2["cluster_rank"].min(),
                                    df_result_good_distil_2["cluster_rank"].max() + 1,
                                    10,
                                )
                            },
                            tooltip={"always_visible": True, "placement": "bottom"},
                        ),
                        html.Button("Reset Cluster Rank", id="reset-cluster-button", n_clicks=0),
                    ],
                    style={"width": "50%", "margin": "auto"},
                ),
                html.Div(
                    [
                        html.P("Show Cluster Labels:"),
                        dcc.Checklist(
                            id="show-cluster-labels",
                            options=[{"label": "Show", "value": "on"}],
                            value=["off"],
                        ),
                    ],
                    style={"width": "50%", "margin": "auto"},
                ),
            ],
            style={"position": "relative", "zIndex": "1001", "marginBottom": "20px"},
        ),
        dcc.Graph(
            id="scatter-plot",
            style={
                "height": "80vh",
                "width": "90vw",
                "position": "relative",
                "zIndex": "999",
            },
        ),
        html.Div(
            [html.Button("Generate LLM Prompt from current selection", id="generate-cluster-button", n_clicks=0)],
            style={"width": "50%", "margin": "auto"},
        ),
        html.Div(
            [
                html.P("Prompt for LLM:"),
                dcc.Textarea(
                    id="cluster-text-output",
                    style={"width": "100%", "height": "200px", "display": "none"},
                    value="",
                ),
            ],
            style={"width": "50%", "margin": "auto"},
        ),
    ]
)


@app.callback(
    Output("scatter-plot", "figure"),
    [
        Input("label-range-slider", "value"),
        Input("cluster-rank-slider", "value"),
        Input("sentiment-dropdown", "value"),
        Input("show-cluster-labels", "value"),
    ],
)
def update_scatter_plot(label_range, cluster_rank_range, selected_sentiment, show_cluster_labels):
    show_labels = "on" in show_cluster_labels

    if selected_sentiment == "gut":
        df_filtered = df_result_good_distil_2
    else:
        df_filtered = df_result_bad_distil_2

    df_filtered = df_filtered[
        (df_filtered["labels"].between(label_range[0], label_range[1]))
        & (df_filtered["cluster_rank"].between(cluster_rank_range[0], cluster_rank_range[1]))
    ]

    outliers = df_filtered[df_filtered.labels == -1]
    clustered = df_filtered[df_filtered.labels != -1]

    fig = px.scatter(
        clustered,
        x="x",
        y="y",
        hover_data=[
            "summary_good_bad",
            "sentiment",
            "cluster_rank",
            "cluster_count",
            "clean_review_br",
        ],
        hover_name="cluster_label",
        color="labels",
        color_continuous_scale="rainbow",
        opacity=0.7,
    )

    if show_labels:
        centroids = clustered.groupby("labels", sort=False).agg(
            {
                "x": "mean",
                "y": "mean",
                "cluster_label": "first",
                "cluster_count": "count",
            }
        )
        for row in centroids.itertuples():
            fig.add_annotation(
                x=row.x,
                y=row.y,
                text=f"{row.cluster_label}, #reviews: {row.cluster_count}",
                showarrow=False,
            )

    fig.add_scatter(
        x=outliers["x"],
        y=outliers["y"],
        mode="markers",
        marker=dict(color="lightgray", opacity=0.5, size=5.0),
        name="No cluster",
        selectedpoints=False,
        hoverinfo="skip",
    )

    fig.update_layout(coloraxis_colorbar=dict(len=0.9, x=1.0), height=600)
    fig.update_traces(marker=dict(size=3), selector=dict(mode="markers"))

    return fig


@app.callback(
    [
        Output("label-range-slider", "marks"),
        Output("label-range-slider", "min"),
        Output("label-range-slider", "max"),
        Output("cluster-rank-slider", "marks"),
        Output("cluster-rank-slider", "min"),
        Output("cluster-rank-slider", "max"),
    ],
    [Input("sentiment-dropdown", "value")],
)
def update_slider_marks(selected_sentiment):
    if selected_sentiment == "gut":
        df_filtered = df_result_good_distil_2
    else:
        df_filtered = df_result_bad_distil_2

    label_marks, min_label, max_label, cluster_rank_marks, min_cluster_rank, max_cluster_rank = (
        preprocess_data_for_slider_marks(df_filtered)
    )

    return (
        label_marks,
        min_label,
        max_label,
        cluster_rank_marks,
        min_cluster_rank,
        max_cluster_rank,
    )


@app.callback(
    Output("label-range-slider", "value"),
    [Input("reset-button", "n_clicks")],
    [State("label-range-slider", "min"), State("label-range-slider", "max")],
)
def reset_label_slider(n_clicks, min_val, max_val):
    return [min_val, max_val]


@app.callback(
    Output("cluster-rank-slider", "value"),
    [Input("reset-cluster-button", "n_clicks")],
    [State("cluster-rank-slider", "min"), State("cluster-rank-slider", "max")],
)
def reset_cluster_slider(n_clicks, min_val, max_val):
    return [min_val, max_val]


@app.callback(
    Output("cluster-text-output", "style"),
    [Input("generate-cluster-button", "n_clicks")],
)
def show_cluster_text_output(n_clicks):
    if n_clicks > 0:
        return {"width": "100%", "height": "200px", "display": "block"}
    else:
        return {"width": "100%", "height": "200px", "display": "none"}


@app.callback(
    Output("cluster-text-output", "value"),
    [Input("generate-cluster-button", "n_clicks")],
    [State("cluster-rank-slider", "value"), State("sentiment-dropdown", "value")],
)
def update_cluster_text_output(n_clicks, cluster_rank_range, selected_sentiment):
    if n_clicks > 0:
        if selected_sentiment == "gut":
            df_text_outp = df_result_good_distil_2
        else:
            df_text_outp = df_result_bad_distil_2

        df_text_outp = df_text_outp[
            (df_text_outp["cluster_rank"] <= cluster_rank_range[1])
            & (df_text_outp["cluster_rank"] >= cluster_rank_range[0])
        ]

        df_text_outp["summary_good_bad"] = df_text_outp["summary_good_bad"].fillna("").astype(str)

        sampled_data = df_text_outp.sample(frac=0.1, random_state=42)

        grouped_data = (
            sampled_data.groupby("cluster_label", sort=False)["summary_good_bad"].agg("\n".join).reset_index()
        )

        prompt_instruction = """Analysiere die nach ### folgenden CLUSTER "Clustertitel", die einzelene Bestandteile von Bewertungen erhalten\nund leite pro Cluster eine Business Massnahme ab um das Hauptproblem des Clusters zu lösen oder zu Verbessern.\nGib das Cluster mit seinem "Clustertitel" sowie die dazugehörige Maßnahme zurück.\n###"""

        cluster_texts = prompt_instruction + "\n\n".join(
            f"\nCLUSTER - {row['cluster_label']}\n{row['summary_good_bad']}" for _, row in grouped_data.iterrows()
        )
        return cluster_texts
    else:
        return ""


if __name__ == "__main__":
    app.run_server(debug=True, host="0.0.0.0", port=7860)