[1]常露予,张晓滨.问答模式下结合属性语义的实体属性抽取研究[J].计算机技术与发展,2024,34(04):174-179.[doi:10. 3969 / j. issn. 1673-629X. 2024. 04. 026]
CHANG Lu-yu,ZHANG Xiao-bin.Research on Entity Attribute Extraction Combined with Attribute Semantics in Question Answering Mode[J].,2024,34(04):174-179.[doi:10. 3969 / j. issn. 1673-629X. 2024. 04. 026]
点击复制
问答模式下结合属性语义的实体属性抽取研究()
《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]
- 卷:
-
34
- 期数:
-
2024年04期
- 页码:
-
174-179
- 栏目:
-
人工智能
- 出版日期:
-
2024-04-10
文章信息/Info
- Title:
-
Research on Entity Attribute Extraction Combined with Attribute Semantics in Question Answering Mode
- 文章编号:
-
1673-629X(2024)04-0174-06
- 作者:
-
常露予; 张晓滨
-
西安工程大学 计算机科学学院,陕西 西安 710048
- Author(s):
-
CHANG Lu-yu; ZHANG Xiao-bin
-
School of Computer Science,Xi’ an Polytechnic University,Xi’ an 710048,China
-
- 关键词:
-
问答模式; 实体属性抽取; 动态注意力; 语义交互; 信息融合
- Keywords:
-
question answering mode; entity attribute extraction; dynamic attention; semantic interaction; information fusion
- 分类号:
-
TP398. 1
- DOI:
-
10. 3969 / j. issn. 1673-629X. 2024. 04. 026
- 摘要:
-
实体属性抽取任务中常面临属性标签过多时模型存在爆炸风险的问题,且目前大多数属性抽取模型对文本均分配一致的注意力因子,未将上下文的变化考虑在内。 为解决上述问题,提出一种基于问答模式的结合属性语义的实体属性抽取方法。 该方法的要点在于,将文本看作上下文,把属性视为查询,从上下文中提取到的答案等同于期望的属性值。文中对文本和属性的语义表示进行建模,并提出一个动态注意力机制用于捕捉二者间的语义交互、实现信息融合,同时自适应地控制属性信息融入文本向量的程度。 为了验证该方法的有效性,将模型与目前广泛应用的 BiLSTM 模型、BiLSTM-CRF 模型、OpenTag 模型和 Open Tagging 模型在包含大量属性标签的数据集 AE-110K、AE-650K 上进行对比实验,结果表明,模型在结合属性语义信息且采用动态 Attention 的条件下,其预测准确度、召回率和 F1 值更高。
- Abstract:
-
Entity attribute extraction tasks often face the problem of model explosion risk when there are too many attribute labels,and atpresent,most attribute extraction models assign consistent attention factors to texts, and do not take context changes into account. Tosolve the above problems,an entity attribute extraction method based on question-answering mode combined with attribute semantics isproposed. The key point of this method is that the text is regarded as the context,the attribute is regarded as the query,and the answer extracted from the context is equivalent to the expected attribute value. The semantic representation of text and attributes is modeled,and adynamic attention is proposed to capture the semantic interaction,realize information fusion,and adaptively control the degree to which attribute information is integrated into the text vector. In order to verify the effectiveness of the proposed method,the model was comparedwith the currently widely used models such as BiLSTM,BiLSTM-CRF,OpenTag and Open Tagging on datasets AE - 110K and AE -650K containing a large number of attribute tags. It is showed that under the condition of combining attribute semantic information and adopting dynamic Attention,the model has higher prediction accuracy,recall rate and F1 value.
更新日期/Last Update:
2024-04-10