be/src/olap/collection_similarity.cpp

// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements.  See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership.  The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License.  You may obtain a copy of the License at
//
//   http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied.  See the License for the
// specific language governing permissions and limitations
// under the License.

#include "olap/collection_similarity.h"

#include "vec/columns/column_nullable.h"
#include "vec/columns/column_vector.h"

namespace doris {
#include "common/compile_check_begin.h"

void CollectionSimilarity::collect(segment_v2::rowid_t row_id, float score) {
    _bm25_scores[row_id] += score;
}

void CollectionSimilarity::get_bm25_scores(roaring::Roaring* row_bitmap,
                                           vectorized::IColumn::MutablePtr& scores,
                                           std::unique_ptr<std::vector<uint64_t>>& row_ids) const {
    size_t num_results = row_bitmap->cardinality();
    auto score_column = vectorized::ColumnFloat32::create(num_results);
    auto& score_data = score_column->get_data();

    row_ids->resize(num_results);

    int32_t i = 0;
    for (uint32_t row_id : *row_bitmap) {
        (*row_ids)[i] = row_id;
        auto it = _bm25_scores.find(row_id);
        if (it != _bm25_scores.end()) {
            score_data[i] = it->second;
        } else {
            score_data[i] = 0.0;
        }
        i++;
    }

    auto null_map = vectorized::ColumnUInt8::create(num_results, 0);
    scores = vectorized::ColumnNullable::create(std::move(score_column), std::move(null_map));
}

void CollectionSimilarity::get_topn_bm25_scores(roaring::Roaring* row_bitmap,
                                                vectorized::IColumn::MutablePtr& scores,
                                                std::unique_ptr<std::vector<uint64_t>>& row_ids,
                                                OrderType order_type, size_t top_k) const {
    std::vector<std::pair<uint32_t, float>> top_k_results;

    if (order_type == OrderType::DESC) {
        find_top_k_scores<OrderType::DESC>(
                row_bitmap, _bm25_scores, top_k,
                [](const ScoreMapIterator& a, const ScoreMapIterator& b) {
                    return a->second > b->second;
                },
                top_k_results);
    } else {
        find_top_k_scores<OrderType::ASC>(
                row_bitmap, _bm25_scores, top_k,
                [](const ScoreMapIterator& a, const ScoreMapIterator& b) {
                    return a->second < b->second;
                },
                top_k_results);
    }

    size_t num_results = top_k_results.size();
    auto score_column = vectorized::ColumnFloat32::create(num_results);
    auto& score_data = score_column->get_data();

    row_ids->resize(num_results);
    roaring::Roaring new_bitmap;

    for (size_t i = 0; i < num_results; ++i) {
        (*row_ids)[i] = top_k_results[i].first;
        score_data[i] = top_k_results[i].second;
        new_bitmap.add(top_k_results[i].first);
    }

    *row_bitmap = std::move(new_bitmap);
    auto null_map = vectorized::ColumnUInt8::create(num_results, 0);
    scores = vectorized::ColumnNullable::create(std::move(score_column), std::move(null_map));
}

template <OrderType order, typename Compare>
void CollectionSimilarity::find_top_k_scores(
        const roaring::Roaring* row_bitmap, const ScoreMap& all_scores, size_t top_k, Compare comp,
        std::vector<std::pair<uint32_t, float>>& top_k_results) const {
    if (top_k <= 0) {
        return;
    }

    std::priority_queue<ScoreMapIterator, std::vector<ScoreMapIterator>, Compare> top_k_heap(comp);

    std::vector<uint32_t> zero_score_ids;
    for (uint32_t row_id : *row_bitmap) {
        auto it = all_scores.find(row_id);
        if (it == all_scores.end()) {
            zero_score_ids.push_back(row_id);
            continue;
        }
        if (top_k_heap.size() < top_k) {
            top_k_heap.push(it);
        } else if (comp(it, top_k_heap.top())) {
            top_k_heap.pop();
            top_k_heap.push(it);
        }
    }

    top_k_results.reserve(top_k_heap.size());
    while (!top_k_heap.empty()) {
        auto top = top_k_heap.top();
        top_k_results.push_back({top->first, top->second});
        top_k_heap.pop();
    }

    if constexpr (order == OrderType::DESC) {
        std::ranges::reverse(top_k_results);

        size_t remaining = top_k - top_k_results.size();
        for (size_t i = 0; i < remaining && i < zero_score_ids.size(); ++i) {
            top_k_results.emplace_back(zero_score_ids[i], 0.0F);
        }
    } else {
        std::vector<std::pair<uint32_t, float>> final_results;
        final_results.reserve(top_k);

        size_t zero_count = std::min(top_k, zero_score_ids.size());
        for (size_t i = 0; i < zero_count; ++i) {
            final_results.emplace_back(zero_score_ids[i], 0.0F);
        }

        std::ranges::reverse(top_k_results);
        size_t remaining = top_k - final_results.size();
        for (size_t i = 0; i < remaining && i < top_k_results.size(); ++i) {
            final_results.push_back(top_k_results[i]);
        }

        top_k_results = std::move(final_results);
    }
}

#include "common/compile_check_end.h"
} // namespace doris